Challenges and Advances

in Computational Chemistry and Physics 35

Series Editor: Jerzy Leszczynski

Supratik Kar
Jerzy Leszczynski Editors

Current Trends in
Computational
Modeling for Drug
Discovery
Challenges and Advances in Computational
Chemistry and Physics

Volume 35

Series Editor
Jerzy Leszczynski, Department of Chemistry and Biochemistry, Jackson State
University, Jackson, MS, USA
This book series provides reviews on the most recent developments in computational
chemistry and physics. It covers both the method developments and their applica-
tions. Each volume consists of chapters devoted to the one research area. The series
highlights the most notable advances in applications of the computational methods.
The volumes include nanotechnology, material sciences, molecular biology, struc-
tures and bonding in molecular complexes, and atmospheric chemistry. The authors
are recruited from among the most prominent researchers in their research areas. As
computational chemistry and physics is one of the most rapidly advancing scientific
areas such timely overviews are desired by chemists, physicists, molecular biologists
and material scientists. The books are intended for graduate students and researchers.
All contributions to edited volumes should undergo standard peer review to ensure
high scientific quality, while monographs should be reviewed by at least two experts
in the field. Submitted manuscripts will be reviewed and decided by the series editor,
Prof. Jerzy Leszczynski.
Supratik Kar · Jerzy Leszczynski
Editors

Current Trends
in Computational Modeling
for Drug Discovery
Editors
Supratik Kar Jerzy Leszczynski
Chemometrics and Molecular Modeling Department of Chemistry, Physics
Laboratory and Atmospheric Science
Department of Chemistry Jackson State University
Kean University Jackson, MS, USA
Union, NJ, USA

ISSN 2542-4491 ISSN 2542-4483 (electronic)

Challenges and Advances in Computational Chemistry and Physics
ISBN 978-3-031-33870-0 ISBN 978-3-031-33871-7 (eBook)
https://doi.org/10.1007/978-3-031-33871-7

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2023

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
For COVID-19 HEROES [Frontline workers,
Health Care Professionals, First responders,
Researchers and Scientists worked for
Vaccines and Small drug molecules]
Preface

Computer-aided drug design (CADD) approaches are one of the rapidly growing
research areas to minimize the experimental efforts to support and seed up the
drug design and discovery [1, 2]. The discovery time of small drug molecules not
only decreased over the years but also the late-stage drug failure is also reduced
due to the early prediction of absorption, distribution, metabolism, excretion toxi-
city (ADMET) profile. Along with the explanation of molecular basis for the
expected biological response, CADD offers precise prediction of possible deriva-
tives and structural scaffolds that would improve therapeutic activity [3, 4]. CADD
involves a series of computational techniques or in silico approaches which comprise
combinatorial chemistry, quantitative structure–active relationships (QSARs), phar-
macophore modeling, rigid, flexible, covalent, and quantum polarized docking,
molecular dynamics simulation followed by molecular mechanics with generalized
Born and surface area solvation (MM/GBSA) and Poisson–Boltzmann surface area
(PBSA) to perform virtual screening, lead optimization, de novo design, and so forth.
Nowadays, among the major in silico approaches, drug repurposing through compu-
tational chemistry is one of the popular one where scientists discover new uses of
already approved drugs by regulatory agencies for another disease to provide the
quickest possible transition from bench to bedside. CADD through in silico tech-
niques helps in the identification and optimization of new drugs employing leverage
of chemical and biological evidence about targets and ligands using computational
power. Not only that, QSAR and machine learning (ML) models offer removal of
unwanted molecules with undesirable ADMET profile to ease the selection of the
most hopeful candidates [1, 2, 5].
Over the years, CADD has made major contributions to speed up the drug
discovery process through the amalgamation of in silico approaches with experi-
mental efforts. Indeed, several marketed drugs such as indinavir, captopril, dorzo-
lamide, ritonavir, oseltamivir, boceprevir, nolatrexed, tirofiban, imatinib, zanamivir,
and nelfinavir have been identified or optimized with the aid of molecular modeling
techniques [6]. We are extremely hopeful that the number will be increased to mani-
fold, and without any doubt, we can say that we are living in the era of CADD and
artificial intelligence (AI)-based drug design and discovery!

vii
viii Preface

The book includes ten chapters encompassing current and advanced computa-
tional modeling techniques and their real-world application for drug design and
discovery for different diseases covering different therapeutic classes.
Chapter 1 by Chakraborti and S talks about multiple components of structure-
based drug discovery (SBDD), its workflow, and associated challenges. Authors also
provided the possible limitations and how these limitations can be overcome which
is extremely important in drug design and discovery.
Chapter 2 prepared by Khatun et al. deals with the structural biology of class IIb
histone deacetylases (HDACs) and talks about how in silico techniques including the
virtual screening approaches have been implemented to design HDAC6 and HDAC10
inhibitors. Furthermore, the interactions of class IIb HDACs with their inhibitors are
also emphasized comprehensively to offer a detail insight. This chapter presents
knowledge for designing newer class IIb HDAC inhibitors in future.
Chapter 3 by Yadav et al. offers important findings involving computational
modeling of small compounds as multitarget-directed ligands (MTDLs) with poten-
tial anti-AD activity which could afford vital leads for discovering new molecules as
novel AD therapeutics to be used for the management of Alzheimer’s disease (AD).
Chapter 4 by Purohit et al. emphasizes the fundamentals of computer modeling
and discusses the relationship between in silico experiments and viral infections
followed by role of computational model in the development of antiviral agents.
Chapter 5 by Gautam and Kumar recapitulates the experimentally tested antivirals
as well as the in silico approaches to identify inhibitors for Nipah virus which will
be helpful for the researchers in antiviral drug discovery against NiV.
Chapter 6 by Gomatam et al. illustrates an overview of the various computational
strategies that have been reported in the discovery of drugs for HIV. A comprehen-
sive overview of several structure-based and ligand-based computational methods is
presented followed by some notable applications of these methods in the discovery
of novel anti-HIV compounds. Authors also discussed the emergence of powerful
machine learning algorithms which have proved to be useful in the design of new lead
molecules and in the development of theoretical models that can predict resistance
to antiretroviral therapy.
Chapter 7 prepared by Chatterjee et al. discusses severe fever with throm-
bocytopenia syndrome virus (SFTSV) disease and its causative agent, epidemi-
ology, pathogenesis, diagnosis, and recent development in the treatment in form
of identification of potential lead using computational modeling.
Chapter 8 by Benfenati et al. thoroughly discusses computational toxicological
aspects in drug design and discovery, screening adverse effects along with existing
in silico tools or future perspectives.
Chapter 9 by Banerjee and Roy demonstrates the read-across and RASAR tools
and different quality and evaluation metrics associated with this research developed in
the Drug Theoretics and Cheminformatics (DTC) Laboratory and their applications
in prediction of different activity/toxicity endpoints.
Chapter 10 by Kar and Leszczynski summarizes major drug databases covering
drug molecules, chemicals, therapeutic targets, metabolomics, and peptides which
Preface ix

are major resources for drug discovery employing drug repurposing, high throughput,
and virtual screening.
The editors convey their gratefulness to all the authors for their knowledge infor-
mative contributions. Furthermore, we thank the reviewers for their time, expertise,
and fruitful comments to improve the book’s quality. We firmly believe that this
edited book will be helpful to all the early days researcher as well as seasoned ones
in the field of CADD irrespective of discipline the budding researcher and experts in
this specific field.

Union, NJ, USA Supratik Kar

Jackson, MS, USA Jerzy Leszczynski

References

1. Roy K, Kar S, Das RN (2015) Understanding the basics of QSAR for applications in
pharmaceutical sciences and risk assessment. Academic press
2. Roy K, Kar S, Das RN (2015) A primer on QSAR/QSPR modeling: fundamental concepts.
Springer
3. Kar S, Leszczynski L (2020) Open access in silico tools to predict the ADMET profiling of
drug candidates. Expert Opin Drug Discov 15:1473–1487
4. Kar S, Roy K, Leszczynski L (2020) In silico tools and software to predict ADMET of new
drug candidates. In: Silico methods for predicting drug toxicity, Benfenati E (ed). Humana,
New York, NY, pp 85–115
5. Kar S, Sanderson H, Roy K, Benfenati E, Leszczynski J (2022) Green chemistry in the synthesis
of pharmaceuticals. Chem Rev 122:3637–3710
6. Baig H, Ahmad K, Roy S, et al (2016) Computer aided drug design: success and limitations.
Curr Pharm Des 22:572–581
Contents

1 SBDD and Its Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Sohini Chakraborti and S. Sachchidanand
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State
of Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Samima Khatun, Sk. Abdul Amin, Shovanlal Gayen, and Tarun Jha
3 Role of Computational Modeling in Drug Discovery
for Alzheimer’s Disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Mange Ram Yadav, Prashant R. Murumkar, Rahul Barot,
Rasana Yadav, Karan Joshi, and Monica Chauhan
4 Computational Modeling in the Development of Antiviral
Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Priyank Purohit, Pobitra Borah, Sangeeta Hazarika, Gaurav Joshi,
and Pran Kishore Deb
5 Targeted Computational Approaches to Identify Potential
Inhibitors for Nipah Virus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Sakshi Gautam and Manoj Kumar
6 Role of Computational Modelling in Drug Discovery for HIV . . . . . . 157
Anish Gomatam, Afreen Khan, Kavita Raikuvar,
Merwyn D’costa, and Evans Coutinho
7 Recent Insight of the Emerging Severe Fever
with Thrombocytopenia Syndrome Virus: Drug
Discovery, Therapeutic Options, and Limitations . . . . . . . . . . . . . . . . . 195
Shilpa Chatterjee, Arindam Maity, and Debanjan Sen
8 Computational Toxicological Aspects in Drug Design
and Discovery, Screening Adverse Effects . . . . . . . . . . . . . . . . . . . . . . . . 213
Emilio Benfenati, Gianluca Selvestrel, Anna Lombardo,
and Davide Luciani

xi
xii Contents

9 Read-Across and RASAR Tools from the DTC Laboratory . . . . . . . . 239

Arkaprava Banerjee and Kunal Roy
10 Databases for Drug Discovery and Development . . . . . . . . . . . . . . . . . 269
Supratik Kar and Jerzy Leszczynski

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Contributors

Sk. Abdul Amin Natural Science Laboratory, Division of Medicinal and Pharma-
ceutical Chemistry, Department of Pharmaceutical Technology, Jadavpur University,
Kolkata, India
Arkaprava Banerjee Department of Pharmaceutical Technology, Drug Theoretics
and Cheminformatics (DTC) Laboratory, Jadavpur University, Kolkata, India
Rahul Barot Faculty of Pharmacy, The Maharaja Sayajirao University of Baroda,
Vadodara, Gujarat, India
Emilio Benfenati Istituto di Ricerche Farmacologiche Mario Negri IRCCS,
Milano, Italy
Pobitra Borah School of Pharmacy, Graphic Era Hill University, Dehradun,
Uttarakhand, India
Sohini Chakraborti Centre for Targeted Protein Degradation, Division of Biolog-
ical Chemistry and Drug Discovery, School of Life Sciences, University of Dundee,
Dundee, UK
Shilpa Chatterjee Department of Biomedical Science, College of Medicine,
Chosun University, Gwangju, Republic of Korea
Monica Chauhan Faculty of Pharmacy, The Maharaja Sayajirao University of
Baroda, Vadodara, Gujarat, India
Evans Coutinho Department of Pharmaceutical Chemistry, Bombay College of
Pharmacy, Mumbai, India
Pran Kishore Deb Department of Pharmaceutical Sciences, Faculty of Pharmacy,
Philadelphia University, Amman, Jordan
Merwyn D’costa Department of Pharmaceutical Chemistry, Bombay College of
Pharmacy, Mumbai, India

xiii
xiv Contributors

Sakshi Gautam Virology Unit and Bioinformatics Centre, Institute of Microbial

Technology, Council of Scientific and Industrial Research (CSIR), Chandigarh, India;
Academy of Scientific and Innovative Research (AcSIR), Ghaziabad, India
Shovanlal Gayen Laboratory of Drug Design and Discovery, Department of
Pharmaceutical Technology, Jadavpur University, Kolkata, India
Anish Gomatam Department of Pharmaceutical Chemistry, Bombay College of
Pharmacy, Mumbai, India
Sangeeta Hazarika School of Pharmacy, Graphic Era Hill University, Dehradun,
Uttarakhand, India;
Department of Pharmaceutical Engineering and Technology, Indian Institute of
Technology (Banaras Hindu University), Varanasi, Uttar Pradesh, India
Tarun Jha Natural Science Laboratory, Division of Medicinal and Pharmaceutical
Chemistry, Department of Pharmaceutical Technology, Jadavpur University, Kolkata,
India
Gaurav Joshi School of Pharmacy, Graphic Era Hill University, Dehradun, Uttarak-
hand, India
Karan Joshi Faculty of Pharmacy, The Maharaja Sayajirao University of Baroda,
Vadodara, Gujarat, India
Supratik Kar Department of Chemistry, Chemometrics and Molecular Modeling
Laboratory, Kean University, Union, NJ, USA
Afreen Khan Department of Pharmaceutical Chemistry, Bombay College of Phar-
macy, Mumbai, India
Samima Khatun Laboratory of Drug Design and Discovery, Department of Phar-
maceutical Technology, Jadavpur University, Kolkata, India
Manoj Kumar Virology Unit and Bioinformatics Centre, Institute of Microbial
Technology, Council of Scientific and Industrial Research (CSIR), Chandigarh, India;
Academy of Scientific and Innovative Research (AcSIR), Ghaziabad, India
Jerzy Leszczynski Department of Chemistry, Physics and Atmospheric Sciences,
Interdisciplinary Center for Nanotoxicity, Jackson State University, Jackson, MS,
USA
Anna Lombardo Istituto di Ricerche Farmacologiche Mario Negri IRCCS,
Milano, Italy
Davide Luciani Istituto di Ricerche Farmacologiche Mario Negri IRCCS, Milano,
Italy
Arindam Maity Department of Pharmaceutical Technology, JIS University,
Kolkata, India
Contributors xv

Prashant R. Murumkar Faculty of Pharmacy, The Maharaja Sayajirao University

of Baroda, Vadodara, Gujarat, India
Priyank Purohit School of Pharmacy, Graphic Era Hill University, Dehradun,
Uttarakhand, India
Kavita Raikuvar Department of Pharmaceutical Chemistry, Bombay College of
Pharmacy, Mumbai, India
Kunal Roy Department of Pharmaceutical Technology, Drug Theoretics and Chem-
informatics (DTC) Laboratory, Jadavpur University, Kolkata, India
S. Sachchidanand Department of Bioinformatics, Zydus Research Centre, Ahmed-
abad, India
Gianluca Selvestrel Istituto di Ricerche Farmacologiche Mario Negri IRCCS,
Milano, Italy
Debanjan Sen Department of Pharmaceutical Technology, BCDA College of
Pharmacy & Technology, Hridaypur, Kolkata, India
Mange Ram Yadav Centre of Research for Development, Parul University, Vado-
dara, Gujarat, India
Rasana Yadav Faculty of Pharmacy, The Maharaja Sayajirao University of Baroda,
Vadodara, Gujarat, India
Chapter 1
SBDD and Its Challenges

Sohini Chakraborti and S. Sachchidanand

Abstract Proteins are the important biological macromolecules that are targeted by
most of the existing drugs. SBDD play a critical role in design of drug-like, novel,
potent, and safe modulators. It is a joint effort from structural biologists and compu-
tational scientists, which considers various limitations of the techniques and suitably
guides drug designers. Identifying a novel, potent, and safe drug-like molecule is a
long challenging path, and throughout this discovery journey, SBDD provides crucial
guiding light at different stages. SBDD involves the use of structural data of target
proteins to identify suitable ligand candidates that might bind the protein of interest
and modulate its functions, resulting in therapeutic benefit. In this chapter, we provide
an overview of computational SBDD workflow, and the various challenges associ-
ated with it. We also discuss strategies that could be adopted to tackle the challenges
by making the best use of available information.

Keywords Structure-based drug discovery (SBDD) · Structure selection · Ligand

screening · Binding affinity prediction · Protein flexibility

1.1 Introduction

The biological functions carried out by a living cell are governed by complex
molecular recognition defined mostly by various non-covalent interactions among
biological macromolecules and small molecule—macromolecule complexes that are

Dedicated to the memory of Late Professor N. Srinivasan

S. Chakraborti (B)
Centre for Targeted Protein Degradation, Division of Biological Chemistry and Drug Discovery,
School of Life Sciences, University of Dundee, 1 James Lindsay Place, Dundee DD1 5JJ, UK
e-mail: [email protected]
S. Sachchidanand
Department of Bioinformatics, Zydus Research Centre, Ahmedabad 380058, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 1

S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_1
2 S. Chakraborti and S. Sachchidanand

present in a cellular milieu [1]. Though the origin of molecular recognition is micro-
scopic in nature, its consequences are macroscopic. The fundamental basis for molec-
ular recognition is inherent in the potential energy surface represented by the inter-
action energy of two or more molecules as a function of their mutual separation
and orientation. The feasibility and strength of any molecular recognition event are
dictated by the extent to which the three-dimensional (3D) structures of the inter-
acting partners complement each other in their shapes and electrostatic features,
given the coherence in their spatiotemporal existence [2]. It is important to empha-
size here that molecular recognition in aqueous biological system is complex with
respect to its predictabilities towards binding while mystery like the role of water in
binding still to be resolved. Proteins, carbohydrates, and nucleic acids are the impor-
tant biological macromolecules that maintain the life processes by interacting with
diverse binding partners called ligands. This chapter would focus on the interactions
between proteins and small molecules (drug-like ligands).
Proteins play a versatile role in maintenance of life and their 3D structures influ-
ence their interactions and functions. Factors influencing changes in the 3D struc-
ture of proteins could eventually alter cellular functions due to changes in their
interaction profiles with their interactomes, which ultimately might lead to disease
phenotypes. Therefore, understanding the 3D structure of proteins (static as well as
dynamic) is of great importance to investigate the cause of the disease at a molec-
ular level [3]. Such understanding guides the rational design of therapeutic agents
(drugs) that can be targeted against the protein of interest to modulate its function
and obtain the desired pharmacological response—the approach being termed as
Structure-Based Drug Design (SBDD). SBDD approaches employ a collection of
computational techniques which provide insights from the 3D static structure of a
protein and its complexes, study their dynamic behaviour at atomic level to guide the
design of modulators targeted against the protein of interest. While SBDD approaches
are routinely used in any drug discovery program, the ability to interpret and ratio-
nally use the multiple layers of information that these approaches provide would
ultimately lead to designing of better modulators. SBDD played an important role in
the approval of the HIV-1 protease inhibitors in the 1990s [4], and gave major boost
to the approach. Since then, structure-based approaches contributed to the approval
of several new drugs in different therapeutic areas [5].
The structural information of the target protein helps in optimizing its interactions
with the potential drug candidates and thus guides in improving its potency [6, 7]
(Fig. 1.1). Not only the structure of the target protein but also the 3D structure
of related proteins (homologs) also play an important role in SBDD by guiding
to achieve specificity in interactions [8] (Fig. 1.1), for example, design of JAKs
inhibitors [9, 10]. Such specificity in interactions contributes to designing of a safer
drug candidate which is a crucial objective in drug discovery. Though serendipity
and high throughput screening (HTS) play an important role in drug discovery and
design, SBDD approach despite its limitations provides a rational foundation towards
discovery of new drugs. Structure-based virtual screening has evolved with the power
of improved and sophisticated computational resources to explore larger chemical
space and provide initial hits that are validated by screening of much smaller set of
1 SBDD and Its Challenges 3

Fig. 1.1 a The Ligand A fits well in Protein A (target) and Protein B (off target) due to similar
binding sites. b The modified Ligand B fits well in Protein A but not in Protein B. The modification
of the ligand helped it to fit better in sub-pocket II of Protein A that is absent in Protein B

test molecules through wet lab experiments, therefore, resulting in better hit rates
than HTS [11–13].
In this chapter, we present an overview of the general workflow of computational
SBDD pipeline, and the common challenges associated with it. Learning from our
experiences, we have further discussed various strategies that could be adopted to
tackle such challenges and making the best use of available information.

1.2 Overview on Structure-Based Drug Design (SBDD)

SBDD is the outcome of collaboration between structural biologists and compu-

tational chemists, who integrate their respective techniques to derive under-
standing from structural models of protein–ligand complexes. Together with medic-
inal chemists and other drug discovery scientists, the understanding translates to
designing novel and potent ligands against the therapeutic target protein of interest.
The main objective of the SBDD approach is to understand the binding pose (confor-
mation and orientation) of a ligand molecule in the protein binding site (an active
or allosteric site). The bioactive conformation of the ligand in the protein binding
pocket (determined either by experimental techniques or predicted by computa-
tional techniques) provides molecular insights to design and optimize potent New
Chemical Entities (NCEs). The insights derived from the structural data primarily
involve probing the shape and the electronic complementarity of protein–ligand
complexes. The complementary stereoelectronic features of protein and ligand
are governed by favourable thermodynamic parameters through various types of
intra- and intermolecular interactions, such as hydrogen bonding, π-π stacking,
hydrophobic contacts [2, 14]. Therefore, binding of a ligand to a protein is the result of
4 S. Chakraborti and S. Sachchidanand

energetic gain from the establishment of multiple interactions. SBDD strategies allow
the optimization of protein–ligand interactions to improve potency and selectivity
by preserving and optimizing selected drug-like properties. Validating predictions
with experimental studies and implementing the learning to improve computational
models are crucial to any SBDD program. Using feedback from various experiments,
SBDD is also useful in designing multi-parameter optimized molecules that helps in
understanding the structure activity relationships (SAR). This understanding can aid
in resolving various ADMET (absorption, distribution, metablolism, excretion, and
toxicity) issues. Optimization of ADMET properties using structural data is another
important segment of computational SBDD as elaborated elsewhere [8, 15].
In this chapter, we have focussed on the key aspects of SBDD that involve ligand
screening and optimization. The basic requirements to initiate a SBDD program are:
(i) availability of a ‘suitable’ protein structure, (ii) 3D structure of ligand molecules,
and (iii) a tool to predict the ligand binding pose in the protein pocket of interest.
Each of these essential requirements for SBDD is discussed below.

1.2.1 Protein Structure

A SBDD approach would require a 3D structure of the apo-/holo-protein that can be

obtained either by experimental or computational means. The availability of ‘good
quality’ (as discussed in Sect. 1.3.1) structural data of the target protein, preferably
bound to a known binder (e.g. endogenous ligand, cofactor, known inhibitor), is
important for SBDD. Such information helps to decipher the location, shape, and
composition of the binding sites (specific regions) on a protein structure that could
mediate its interactions with potential drug molecules and modulate the desired
biological functions. The size, volume, degree of preorganization (conformational
flexibility/rigidity), polarity, and polarizability analysis of the binding sites are very
useful. The features of the binding site of interest are therefore the guiding cues to
identify agents that engages in interactions with critical residues of the target protein
and are a good fit in terms of shape and electrostatic features [15]. The residues
critical for obtaining desired biological response could be identified from various
sources such as mutational data [16], pre-existing SAR of modulators [17], or could
be anticipated from bioinformatic analyses of protein sequence conservation [18].
The publicly available structural repository, the Protein Data Bank (PDB) [19],
holds more than 190,900 (as of September 2022, https://www.rcsb.org/stats/sum
mary) structures of biological macromolecules that contain at least one protein
molecule. In this section, we briefly discuss various techniques for obtaining 3D struc-
tural data of proteins that are essential pre-requisites for a SBDD pipeline. Readers are
encouraged to refer to the cited literatures for details of each technique that is beyond
the scope of discussion in this chapter. Three experimental techniques primarily
contribute to providing atomic coordinates of protein and protein–ligand complexes.
These are macromolecular X-ray crystallography (MX) [20], nuclear magnetic reso-
nance (NMR) [21], and cryogenic electron microscopy (Cryo-EM) [22] with each
1 SBDD and Its Challenges 5

having its own advantages and limitations. Computationally, the structure can either
be generated by homology modelling/comparative modelling [23] (if possible) or
can be extracted from the AlphaFold database [24] (if a ‘suitable’ structure is avail-
able). The quality of structural information derived from protein models predicted by
classical threading-based methods (TBMs) [25] is not suitable for SBDD. However,
an improved TBM combined with structural approaches, FINDSITEcomb , has been
demonstrated to identify potential binders of a target protein, and its performance
to a great extent is shown to be insensitive to structural quality [26]. In the absence
of any suitable structure of the target protein, the FINDSITEcomb technique could
be useful during early drug discovery stage for identification of potential in silico
hits from large chemical space. Advanced ligand design and lead optimization stages
demand high-quality of structural data, and hence, the current TBMs are unlikely to
provide reliable information.

1.2.1.1 Experimental Protein Models

MX involves crystallization of the protein molecules that leads to locking the protein
in a single conformation within a crystal lattice. Under physiological conditions,
proteins ‘jiggle’ and ‘wiggle’ to perform their functions [27]. Therefore, in its native
state, a protein exists as an ensemble of structural conformations. It could be possible
that the conformation of the protein or its specific regions as trapped during crystal-
lization is not relevant to native biological conditions. This could be, especially, for
the regions of the protein that have higher flexibility or when the crystallization condi-
tions are far away from the physiological environment of the target protein [28, 29].
Further, determining high-resolution protein structures bound to small molecules,
that is a desirable pre-requisite for SBDD, is a challenging task and errors in struc-
tural data are not uncommon [30, 31]. Proteins that lack stable secondary structures
due to higher flexibility are disordered and are generally not amenable to MX—a
problem that can be overcome with NMR as the latter allows capturing the struc-
tural information of a protein in solution state (wherever size of the protein is not a
limitation to NMR). NMR spectroscopy is a very useful tool that not only provides
the structural information of the protein in solution phase but can also help in under-
standing the dynamics of a wide range of biological macromolecules and hence can
provide better functional insights than MX. However, NMR is currently restricted
to small and medium-sized proteins [32]. The technique has limitations with respect
to the speed, and size of molecules that can be tackled when compared to X-ray
diffraction methods (for well-diffracting crystals). Wherever it is possible, NMR
spectroscopy facilitate atomic resolution studies of sparsely populated, transiently
formed biomolecular conformations that exchange with the native state [33]. The
dynamics of the macromolecules and their complexes can also be studied when only
X-ray structure is available, using computational techniques like molecular dynamics
simulations (as discussed in Sect. 1.3.5). Cryo-EM has undergone a ‘resolution revo-
lution’ in the recent years and is rapidly emerging as an important tool for SBDD
[34]. This technique has the potential to capture structures of large macromolecular
6 S. Chakraborti and S. Sachchidanand

assemblies in near-native conformations. Currently, the application of Cryo-EM is

mostly restricted to large proteins/protein assemblies with less success for smaller
ones. By achieving atomic resolution using Cryo-EM, the structural assembly of
proteins and their complexes can be understood that cannot be easily examined by
X-ray crystallography [35].

1.2.1.2 Computational Protein Models

In the absence of an experimental protein structure, computational techniques that

help to predict 3D structures of proteins, such as homology modelling, can guide
SBDD approaches [36]. Such modelling techniques use the known structural infor-
mation (template) of related proteins (homologs) to predict 3D structures of the
target protein. The artificial intelligence (AI)-based methods like AlphaFold [37]
and RosTTAFold [38] have shown remarkable success in the recent times. Albeit
these methods do not require structural information of related proteins for predicting
structures of target proteins, but the prediction algorithms are trained on the existing
structural data available in the PDB. Hence, it is likely that success of these methods
would be higher for protein classes and their conformations that are well repre-
sented in the PDB [39]. Studies have demonstrated that refinement of computational
protein models by molecular dynamic simulations to generate suitable conforma-
tional states improves the accuracy of the models and hence their predictability
efficiencies [40, 41].
Once the protein structure is in place, the second input for SBDD studies is the
availability of drug-like libraries and collection of potential analogues/design ideas
for screening. SBDD can help in screening of the physical ligand libraries based
on experimental 3D coordinates of protein–ligand complex [42, 43], and/or it can
facilitate virtual screening of in silico library of compounds to identify hits [44].
The former is time taking as it involves experimental methods, whereas the latter
is comparatively faster and require lesser investment of resources. The validated
hits identified from screening would then move through various phases of SBDD to
obtain lead compounds that are then optimized.

1.2.2 Ligands

Besides protein structure, the 3D structures of ligand molecules are the other impor-
tant inputs for SBDD. The ligand library intended for screening could be a small
number of analogous compounds or a large set of drug-like molecules that preferably
follow rule of five (Ro5) [45] and passed through pan-assay interference compounds
(PAINS) [46] and rapid elimination of swill (REOS) [47] structural filters. Screening
large and diverse set of molecules against the target ensures maximum coverage of
the chemical space and helps to identify multiple starting points with diverse scaf-
folds for hit to lead generation stage. The objective of the SBDD program influences
1 SBDD and Its Challenges 7

the design of the input ligand library. The main objectives that guide library design
are: (i) hit generation, (ii) fragment (MW < 300 Da) identification for fragment-
based drug design (FBDD), (iii) hit to lead generation, and (iv) lead optimization.
The discovery libraries are designed to address the first objective, i.e. hit genera-
tion. Fragment library is designed to identify fragments to be linked using FBDD
approach to design and identify hits. Fragment-based design requires screening frag-
ments (MW < 300 Da), which are not intrinsically drug-like, but become fragments
of drug-like compounds upon combining [48]. The next goal of library design is to
identify a lead against a target. Such libraries are known as focussed library, which
are built around certain structural motif (known to be active against the target of
interest) or against identified pharmacophoric features known to be important for
binding [49].

1.2.2.1 Library Design

A library is a collection of already synthesized/synthesizable compounds or frag-

ments that could be screened against the therapeutic target. The chemical space and
its diversity accounted for within the library are inversely proportional to the amount
of information available for the target’s binding site. Selection or design of library is
dependent on its intended use. For example, if library is required for limited target
classes, a focussed library would serve the purpose. The clustering density (degree
of structural similarity of library members) must be high where repetitive screen for
similar targets is desired while for diverse targets, a library with lower density would
ensure maximum degree of diversity in their collection. Chemoinformatic tools play
significant role in designing compound library for an identified therapeutic target
[50]. Figure 1.2 summarizes several stages involved in library design.

Fig. 1.2 Examples of the steps involved in designing compound library for virtual screening, a by
building a diverse set drug-like library, b additional steps involved in building a focussed drug-like
library
8 S. Chakraborti and S. Sachchidanand

1.2.2.2 Ligand Screening/Optimization

Upon obtaining satisfactory quality of 3D structural information of target protein

and drug-like molecules, the next step in SBDD is in silico screening/optimization.
Screening of physical stock of millions of compounds is not only practically chal-
lenging but also poorly rewarding. To accelerate this process, computational methods
play a crucial role in predicting potential binders and their binding modes. Predicted
potential binders are then evaluated using various biochemical/biophysical/cell-
based assay methods to identify and rank order the validated hits. A good binder(s) is/
are selected for 3D structure determination in complex with the target protein which
helps in validating the binding pose of the compound/s obtained by computational
studies and understanding SAR for further optimization of the initial binders.
In drug discovery, multiple cycles of optimization of a lead are carried out (without
significant compromise on affinity towards the target) before declaring a candidate.
For example, the designing of ligands can be done to complement the binding site
features (viz., shape and electrostatics) of target and optimization would further fine
tune intermolecular interactions and steric complementarity with the binding site for
achieving optimum thermodynamic parameters. Every cycle of optimization, be it
improving metabolic stability [51] or ADME properties [52] or Cyp liabilities [53],
requires a new set of SAR, involving new designs of molecules around the lead.
It is important to emphasize here that there is a rule of ‘no rule’ to design drugs.
Various tips and tricks from past examples and experiences might sometime work
rationally and sometime serendipitously. Therefore, reinforcing methods with other
techniques and integrating knowledge from various reliable sources are advantageous
in any drug discovery program.

1.2.3 Molecular Docking Simulations

Molecular docking is one of the most popular computational techniques that is

routinely applied in SBDD to gain first insights into plausible design hypotheses
[54]. Molecular docking studies help in identifying compounds against the target
protein through predicted docking pose at the binding site of interest. The quality
of pose is assessed using docking score. Molecular docking analysis also provides
enrichment of huge compound library (enrichment from docking is its ability to rank
large proportion of the active compounds at the top of the proposed list for exper-
imental evaluation) for screening only a few hundreds of compounds by a process
called structure-based virtual screening. Suggested binding mode of ligand through
docking is useful in optimizing interactions of the compound with the target protein
and hence help in improving potency and selectivity.
Any docking method would require the 3D structural information of protein target
and ligands. As discussed later in Sect. 1.3, the docking method has its own limita-
tions, and success of this technique depends on several factors, for example, quality
of protein and ligand structures, identifying correct ionization profile of binding site
1 SBDD and Its Challenges 9

residues, nature of binding site (rigid vs. flexible), selection of force field and scoring
functions. Over the past few years, numerous artificial intelligence/machine learning
(AI/ML)-based techniques have also been reported that predict compound binding
affinity and binding modes with greater accuracies than existing methods [55–58].
However, one of the major challenges of AI/ML-based techniques is its dependence
on availability of large datasets that are required to train most of these algorithms
[5, 59]. Therefore, in unique and novel cases with limited data, AI/ML methods are
unlikely to make meaningful predictions.

1.3 Crucial Components and Challenges in Computational

SBDD

The success of any computational SBDD program is largely dependent on the extent
to which the drug-target interactions (as it happens under physiological conditions)
could be mimicked in silico. Realistic representation of the biophysical events within
the computational algorithm is likely to result in accurate predictions. Unfortunately,
the improvement in accuracy of predictions comes with the compromise in speed
of the calculations. To strike a balance between speed and accuracy, it is neces-
sary to incorporate approximations in the algorithms. It is important that depending
on the question to be addressed and availability of resources, one should adopt the
appropriate strategy at each stage in the computational drug discovery pipeline. In
the following paragraphs, we discuss about the crucial components in computa-
tional SBDD workflows, associated challenges, and ways to handle such challenges
(Fig. 1.3).

Fig. 1.3 Various components in computational SBDD. P Protein, L Ligand, PL Protein–Ligand

complex
10 S. Chakraborti and S. Sachchidanand

1.3.1 Target Structure Selection

The foremost step in computational SBDD is to select the appropriate structure/s

of the target protein that will be subsequently used for compound screening and
optimization. The success of ligand screening/optimization is greatly dependent on
the quality of the input protein structure. High-quality crystal structures of protein–
ligand complex provide the platform to generate a sound design hypothesis. There are
many literatures discussing various components of SBDD that we have highlighted
in Sects. 1.3.2–1.3.5. However, to the best of our knowledge, there are hardly any
published literature that provides a comprehensive guide to aid decision-making for
selecting the suitable starting structure/s. This is mostly because structure selection
for computational SBDD is dependent on several factors that are beyond generaliza-
tion. Here, we have aimed to discuss various factors that could help the beginners in
the field to form an idea about the rational thought process that generally influences
input structure selection. In an ideal situation, a SBDD project demands the prior
availability of multiple high-quality structures of the apo- and holo-protein of interest.
However, it is difficult to satisfy all desired criteria for structure selection and rational
judgement needs to be applied to select the best suitable inputs. Table 1.1 presents a
few hypothetical scenarios that may aid decision-making to choose a suitable struc-
ture in certain practical scenarios. We emphasize that these are only guidelines and
case-specific decisions could be influenced by several practical limitations.

1.3.1.1 Quality of Structure

While ‘resolution’ of X-ray structure is one of the common indicators of its quality,
it is not always directly related to the accuracy of data. Resolution is a measure of the
quantity of data collected and not its quality. While experienced researchers might
be aware, it might not be obvious to novice users that high-resolution structural data
need not always guarantee the reliability of local structural data such as protein–
ligand binding sites [60]. It is recommended to use the combination of Rfree and the
diffraction component precision index (DPI) of a structure instead of its resolution to
get a better impression about the overall model quality and hence the reliability of the
atomic positions within that model [61, 62]. There are several reports that emphasize
errors in high-resolution structural data is not uncommon, and it is advised to verify
the co-ordinates against the experimental evidence, such as electron density maps in
case of crystal structures [31, 63]. Quality parameters such as real space correlation
coefficient (RSCC) [64] and electron density scores for individual atoms (EDIA)
[65] quantify the electron density fit of the structural entity and are useful indicators
of quality of coordinates in local regions of a structure. RSCC ≥ 0.9 and EDIA ≥
0.8 indicate good fit of the co-ordinates with experimental data. The RCSB PDB
has recently introduced the ligand quality slider feature and included assessment on
experimental data fitting and geometry of bound ligands of interest (https://www.rcsb.
org/docs/general-help/ligand-structure-quality-in-pdb-structures). This is one of the
1 SBDD and Its Challenges 11

Table 1.1 Guide for target structure selection for virtual screening using SBDD approach to identify
potential hit compounds against target Protein X that has more than one structure available as
a potential starting point. Note that mutation of an amino acid residue in Protein X causes Disease
Y and this mutation locks Protein X in an inactive conformation
Parameters Structure 1 Structure 2 Remark
Quality: 1.8 Å 2.5 Å Though Structure 1 has better
(a) Resolution 0.45 0.98 resolution, the electron density fit of
(b) Ligand RSCC the ligand as indicated by RSCC is
poor (See Section 1.3.1.1). So,
Structure 2 should be preferred.
However, visual inspection of the
ligand pose in the crystal structure
against the electron density map is
recommended
Sequence Wild type Disease relevant Structure 2 should be preferred as it has
mutation the relevant mutation in its sequence
Ligand bound/ Bound Bound (inhibitor) As the study would aim to design an
unbound (activator) inhibitor, so an inhibitor bound
structure, i.e. Structure 2 should be
preferred
Conformational Active Inactive The aim of the project is to target the
state inactive state of the protein of the
interest. Hence, Structure 2 should be
preferred
Crystallization 3.5 6.8 Selection of the target protein structure
condition pH should be based on the pH of its
environment under physiological
condition

easiest ways to quickly verify the quality of the ligand bound to protein of interest that
is deposited in the PDB. In the absence of target protein structure with satisfactory
quality in the PDB, the PDB-REDO [66] database could be searched to check for
the availability of a better quality model.

1.3.1.2 Sequence Information

Examining the amino acid sequence of the target protein is another important compo-
nent of structure selection. Depending upon whether targeting the wild type or a
mutated protein (such as in many cancers [67]) is of interest, care should be taken
to choose the appropriate starting structures. A single change of amino acid can
appreciably alter the structure of a protein and hence affect its binding with its part-
ners (especially when such change is in the binding site). If a suitable structure of
the target protein is not available, the common practice is to use the structure of a
close homolog. While choosing a structural homolog, it should be ensured that the
binding site features are largely conserved between the target protein and the chosen
12 S. Chakraborti and S. Sachchidanand

homolog so that there are minimal chances of interference with prediction outcomes.
Computational tools like ProBis [68] are helpful to compare the structural similarity
of protein–ligand binding sites.

1.3.1.3 Apo/Holo Conformation

The ligand bound state (holo) of the protein is convenient to use for SBDD when the
screened ligands are intended to bind to the known site. The apo state structure of
a protein does not provide the information on ligand binding site, unless otherwise
any other experimental evidence exists. Computational tools such as Fpocket [69],
SiteMap [70] are useful to predict potential druggable sites (regions amenable to
functional modulation upon binding of drug molecules) on a protein structure. The
ligand bound (holo) or unbound (apo) state of the protein structure may influence
the screening outcomes. It has been shown earlier that holo state structure gives
better enrichment compared to apo state structures as the protein binding site in the
former is already preformed to accommodate similar ligands [71]. It is known that
ligands with different size and/or belonging to different chemical classes may trigger
varying conformational changes in the protein binding site—a phenomenon termed
as induced fit [72]. If multiple experimental structures of the target protein bound to
different chemical classes of ligands show significant conformational changes in the
binding site, it is worth considering an ensemble of structures that are representatives
of each chemical class of ligands. Such an approach would avoid bias and minimize
the chances of missing promising hits that might prefer one conformation over the
other. Also, a ligand that is predicted to bind to majority of the conformers in the
structural ensemble would have higher likelihood of binding to the target protein.
However, sometimes the screening program might be intended to identify new ligands
that have pharmacophores similar to a particular known binder [73]. In such scenarios,
it is justified to use only the protein structure that is bound to a ligand with desired
pharmacophoric features rather than using an ensemble approach. In the absence of
appropriate holo state structures, induced fit docking [74] and molecular dynamics
approaches [75], as discussed later, could be helpful to obtain suitable conformation
of the target protein that can then be used as starting structure/s.

1.3.1.4 Effect of Neighbouring Residues

It could be possible that the ligand binding site of interest lies at or proximal to
protein–protein interfaces. Under biological conditions, these interfaces could be
formed by homomers or heteromers (e.g. in a multi-protein complex). In a slightly
different scenario, one may encounter a situation where the structure of only a single
domain from a multi-domain protein target of interest is available. If the ligand
binding site is close to domain–domain interface, it could be possible that binding of
a ligand to one domain is affected by the contribution of residues from a neighbouring
domain. Such scenarios require careful assessment and if possible, multiple chains/
1 SBDD and Its Challenges 13

domains that form the ligand binding site should be preferred to account for the
contribution from all binding site residues during the ligand binding event.

1.3.1.5 Flexibility Signatures

Dealing with flexibility signatures of ligand binding site is crucial to structure selec-
tion in computational studies. Higher flexibility of a protein residue could be mani-
fested as multiple conformations of its side chain, higher B-factors, and missing
regions in the electron density maps of crystal structure [27]. In case of multiple
conformations, the common practice is to consider the one that has the highest occu-
pancy value. In our experience, it is worthy to consider each of the multiple confor-
mations of the side chains as separate input structural models for ligand screening/
optimization in order to be closer to realistic conditions. This also enhances the
chances of identifying promising hits that prefer to bind to one of the conformers
more strongly than other as mentioned earlier in Sect. 1.3.1.3. A structure that has
missing residues in its ligand binding site could not be used for the prediction of ligand
binding mode and its binding affinity towards the target using SBDD approaches. If
no other suitable structure of the target protein is available, it is necessary to build
the missing residues using appropriate tools like Modeller [76], Prime [77].

1.3.1.6 Protein Conformational State

Proteins can sample a multitude of conformational states in the energy landscape.

These different conformational states are often associated with distinct biological
functions mediated by the protein [27, 78]. Kinases, the popular drug targets, are
known to exist in multiple conformations [79]. The binding site geometry of the active
state of kinases is known to be more conserved than the inactive state conformations
among all kinases [80]. Thus, designing conformation-specific kinase targeted drugs
could help in avoiding undesired effects [81–83]. Again, post-translational modifi-
cation of a protein under biological conditions may affect its conformation and thus
may alter drug binding capacity [84]. It is, therefore, essential to ensure that the
input structure selected for SBDD represents the conformation of the protein that
is intended to be targeted. If such a structure of the target protein or its homolog
is unavailable, molecular dynamics and other conformational sampling techniques
could be employed to predict the conformations that are likely to be closer to the
physiological conditions [85].

1.3.1.7 Unavailability of Target Structure

The structure selection criteria discussed so far assume that experimental structures
of the target proteins are available. However, determining structures of many targets,
14 S. Chakraborti and S. Sachchidanand

such as membrane proteins and intrinsically disordered proteins, are highly chal-
lenging. Computational models could be used in the SBDD programs intended to
target proteins for which experimental structures are unavailable (as mentioned in
Sect. 1.2). Scenarios where an experimental structure of the target protein or its close
homologue is known but is unsatisfactory for SBDD would also require computa-
tional models as the starting point. It is important that any predicted structural model
of protein used in SBDD should satisfy the quality evaluation [86] and represent
the conformational state intended to be targeted by the designed modulator. Detailed
discussion of computational structure prediction methods is beyond the scope of this
chapter but can be found elsewhere [87].

1.3.2 Target and Ligand 3D Structure Preparation

The structure of the target protein obtained from the PDB, and the ligand structures
obtained from chemical libraries are generally not suitable for computational studies
as it is. These structures require pre-treatment, as discussed below, to fix certain
issues and obtaining reliable predictions from the models.

1.3.2.1 Protein Structure Preparation

A typical PDB file of protein and/or protein–ligand complexes might not contain
all the information that are required for initiating any modelling studies. The co-
ordinates of hydrogen atoms are generally absent in the macromolecular crystal
structures, unless the structure is of ultra-high resolution [88]. It is important that the
hydrogen atoms are added to the protein structure before using them for any SBDD
applications. It should also be ensured that the added hydrogen atoms have the right
geometry to optimize the local hydrogen bonding network and the final structure used
as input should be free of steric clashes. This might sometimes require flipping the
side chain of certain residues like histidine, asparagine, and glutamine. Assignment
of protonation and tautomerization states of the binding site residues (especially His,
which can be neutral with a proton either on Nδ or Nε or have a positive or negative
charge) play a critical role. Incorrect ionization would interfere with docking scores
and hence affect the rank order of the screened compounds. The issues with missing
atoms as mentioned in Sect. 1.3.1.5 need to be fixed, and assignment of right charges
to the protein residues at the desired pH should be ensured to mimic the biological
conditions. Removal of co-ordinates of water molecules from the binding site of a
protein crystal structure before docking simulation is generally recommended unless
there is enough experimental evidence to believe that such water molecules play an
important role in protein–ligand interaction. Freely available tools through the WHAT
IF web interface [89] or paid tools like Protein Preparation Wizard [88] available
through Schrodinger suite are few of the many computational tools that could be
used for preparing the input protein structures.
1 SBDD and Its Challenges 15

1.3.2.2 Ligand Structure Preparation

The 2D or 3D structures of ligands could be obtained from publicly available chem-

ical libraries such as PubChem [90], ChEMBL [91], BindingDB [92]. The 2D
ligand structure either downloaded from chemical databases or drawn using chem-
ical sketchers (like ChemDraw [93]) requires conversion into its 3D form for any
SBDD studies. The 3D structures that are available from these libraries may not be
suitable for direct use as it would require geometry optimization in a manner that
is compatible with the force field [94] to be used in the subsequent steps. Further,
assignment of bond orders, and charges, generation of right tautomer and ionization
states of the ligands are essential prior to usage to represent the biological condi-
tions. The library molecules must be passed through different filters (as discussed
in Sect. 1.2.2) which are implemented to build drug-like/lead-like/fragment library.
Apart from these filters, library molecules must also be filtered to exclude those that
have toxic or metabolically unstable groups or molecules known to be chemically
reactive which can interfere with assays (refer Sect. 1.2.2) [46, 47]. One of the impor-
tant aspects of ligand preparation when it contains a chiral centre is to consider the
stereoisomer that is relevant for biological activity. The information on bioactivity
of different stereoisomers of a given ligand against the target of interest could be
obtained from experimental studies. In the absence of such experimental data, it is
wiser to consider all possible stereoisomers for at least the early stages of compu-
tational studies. OpenBabel (freely available) [95] and LigPrep (commercial tool
available with Schrodinger Suite) [88] are two of the many available computational
tools that help in ligand preparation.

1.3.3 Binding Affinity and Mode Prediction

The central focus of computational SBDD programs is: (a) virtual screening and
(b) to predict the binding poses of the ligands in the target protein pocket [96].
The scoring algorithms differentiate between ‘good’ and ‘bad’ binding poses for
individual ligand molecules and provide decent enrichment factors from virtual
screening. Docking score is often wrongly interpreted as a measure of affinity of
a ligand towards the target. Notably, docking scores are generally known to have
poor correlation with experimental binding affinities [97, 98]. The accurate calcu-
lation of binding free energy (△Gbind ) would require exhaustive sampling of the
molecular system in explicit solvent environment both in bound and unbound state.
Since it is time consuming and accurate in silico representation of the physical laws
that govern the event is difficult and complex, several approximations are used to
reduce the complexity of the system [99]. Most docking programs involve sampling
the ligand conformations (with limited degrees of freedom) within the rigid pocket
of the target protein and subsequently ranking the poses based on the goodness of fit
using a scoring function. The scoring functions in most cases are just the approximate
representation of the binding energy that completely or partly neglect the entropic
16 S. Chakraborti and S. Sachchidanand

contributions and only considers the enthalpic component. The enthalpy component
is also a simplistic representation of the protein–ligand interactions. It disregards
the solvent effects and so-called non-classical (for example, CH-π, halogen bonds,
S… O) interactions that are important component of binding energy [100, 101]. We
are probably yet to discover many such ‘non-classical’ interactions and thus far
from incorporating those in scoring functions. Due to these limitations, it is quite
common to encounter cases when the best ranked docked pose as suggested by any
docking algorithm is not the biologically meaningful pose. However, even with the
numerous limitations, docking simulations are undoubtedly one of the most useful
computational tools to distinguish potential binders and non-binders from the vast
chemical libraries in a reasonably less time [102]. With the remarkable advancement
of computing resources in the past few years, employing flexibilities to the receptor
and including solvent contribution to certain degrees even in early stage computa-
tional studies could now be routinely done within reasonable time [103]. Induced fit
docking approaches that allow sampling the side chain conformations of the binding
site residues in the presence of the ligand candidates are shown to improve predic-
tion efficiencies [74]. Other advanced techniques to predict binding affinity and
binding pose such as free energy perturbation (FEP), quantum mechanics/molecular
mechanics (QM/MM), though computationally expensive, when rationally combined
with the traditional rigid docking protocols could be helpful [104–106]. Analysing
the results from any of these prediction approaches in the light of the biological
understanding would ultimately dictate the success of these methods. Ensuring if the
predicted pose of the ligand is engaged in interactions with the functionally impor-
tant protein binding site residues and retain similar interaction fingerprints as that of
known binders are a few strategies to select the meaningful poses from the pool of
suggested docking solutions.

1.3.4 Contribution of Water

Under physiological conditions, the protein and the ligand molecules are solvated.
The binding event between the target protein and the ligand necessitates desolvation
of both the molecules. Desolvating the polar atoms of protein residues or ligand leads
to unfavourable change in enthalpy but facilitates movement of thermodynamically
stable water molecules into the bulk resulting in gaining of entropy [107, 108]. The
degree of disorder of such water molecules prior to their replacement governs the
thermodynamic signature of the water release (hydrophobic effect) and the effect can
range from entropic to enthalpic [109]. It contributes to a favourable enthalpy term
for the hydrophobic contacts in tight pockets and to a favourable entropy term for the
polar interactions. Thus, water molecules play a crucial role in the drug-target binding
event. The thermodynamic signatures of the water molecules in the binding site can
guide in understanding if displacement, replacement, or retention of a particular water
molecule could be helpful for improving potency and/ or achieving specificity [110,
1 SBDD and Its Challenges 17

111]. Explicit modelling of water in protein–ligand binding event is computation-

ally expensive. Therefore, most docking algorithms either neglect the contribution
of water or use a crude representation to account for it. Such a reductionist approach
introduces inaccuracies in predicted binding affinity. Computational techniques that
use implicit water model, like Prime-MMGBSA [112], to predict binding affini-
ties are shown to better correlate with experimental binding affinity for congeneric
series of ligands than traditional docking approaches. Albeit computationally expen-
sive, exploiting thermodynamic signatures of binding site water molecules during
advanced lead optimization phase could prove helpful particularly in scenarios where
water thermodynamics play a dominant role.

1.3.5 Effect of Dynamics

One of the main reasons for the poor correlation between predicted binding affini-
ties obtained from docking studies and experimental binding affinities is due to lack
of consideration of the inherent plasticity of the target protein in the presence of
the ligand [113]. From local rearrangement of side chain and/or backbone atoms of
binding site residues to large-scale movement of loops or domains could be trig-
gered during a protein–ligand binding event [114]. Such movements may dictate
the stability of the interactions between the protein and ligand atoms and hence the
binding affinity. The success rate of computational ligand screening programs that
employ only rigid docking approach for proteins with flexible ligand binding sites
is likely to be low [115]. Molecular dynamics (MD) simulations are useful tools to
predict atomic movements of protein–ligand complexes in explicit solvent environ-
ment [116]. Thus, MD simulation can be used to test and validate the stability of
protein–ligand complexes predicted from docking studies. The information derived
from these simulations could help in incorporating suitable functional groups on the
ligand to maximize its interactions with the protein. Further, MD snapshots could
be used to generate ensemble of structures representing different conformations of
the target protein [117]. The ensemble of structures could then be used to dock the
libraries of ligands and this approach has shown to improve the efficiency of docking
studies [118]. Undoubtedly, MD simulations are computationally expensive but with
advent of graphical processing units (GPUs) and advancement of other computa-
tional resources, it is now feasible to simulate the dynamics of biomolecular systems
at millisecond scales [119, 120]. Depending upon the question that is intended to be
addressed, the right level of flexibility could be introduced into simulation, striking
a balance between accuracy and speed [115].
18 S. Chakraborti and S. Sachchidanand

1.4 Conclusion

In this chapter, we have discussed various components of computational SBDD

and associated challenges. These challenges arise majorly due to limitations with
respect to: (i) computing speed and (ii) accuracy of prediction. The former could
be addressed by advancement in hardware as we are witnessing in the recent times
with the advent of the GPUs [121] and high-performance clusters (HPCs) [122]
era. Strategic investment in computing resources is, thus, essential in any modern-
day drug discovery programs. The latter challenge requires better understanding
of biology and the physical and chemical laws that drive the biological process.
Understanding in detail how molecules interact with each other in physiological and
pathological states is central to modern drug discovery. Accurate representation of
such understanding in computer algorithms by minimizing the approximations is
likely to improve prediction efficiencies.
We would like to emphasize that the possible ways to tackle some of the challenges
as we have discussed in this chapter are based on common scenarios in computational
SBDD. Case-specific issues might demand specialized strategies. In our opinion, the
fundamental strategy to improve the confidence of any prediction is to rationally
combine the use of multiple algorithms that work on varying principles and verify if
there is a consensus in the outcomes. Selection of the best predicted solutions should
be guided by understanding the governing physical and chemical laws and their
biological relevance. Experimental validations are essential to test the correlation
between theoretical and experimental studies. The learning from the experimental
studies should be integrated to the computational pipelines to improve the predic-
tive power of the algorithms. In other words, judicious integration of computational
methods with experimental techniques by considering the limitations of both is one
of the key drivers of any successful SBDD program.

References

1. Alberts B, Johnson A, Lewis J et al (2015) Molecular biology of the cell

2. Patrick G (2018) An introduction to medicinal chemistry (6th edn). Oxford University Press,
Oxford
3. Anderson AC (2003) The process of structure-based drug design. Chem Biol 10:787–797.
https://doi.org/10.1016/j.chembiol.2003.09.002
4. Wlodawer A, Vondrasek J (1998) INHIBITORS OF HIV-1 PROTEASE: a major success of
structure-assisted drug design. Annu Rev Biophys Biomol Struct 27:249–284. https://doi.org/
10.1146/annurev.biophys.27.1.249
5. Batool M, Ahmad B, Choi S (2019) A structure-based drug discovery paradigm. Int J Mol
Sci 20:2783
6. Náray-Szabó G (1993) Analysis of molecular recognition: steric electrostatic and hydrophobic
complementarity. J Mol Recognit 6:205–210. https://doi.org/10.1002/jmr.300060409
7. Yazhini A, Chakraborti S, Srinivasan N (2021) Protein structure, dynamics and assembly:
implications for drug discovery—innovations and implementations of computer aided drug
discovery strategies in rational drug design. In: Singh SK (ed) Springer, Singapore, pp 91–122
1 SBDD and Its Challenges 19

8. Stoll F, Göller AH, Hillisch A (2011) Utility of protein structures in overcoming ADMET-
related issues of drug-like compounds. Drug Discov Today 16:530–538. https://doi.org/10.
1016/j.drudis.2011.04.008
9. Schwartz DM, Kanno Y, Villarino A et al (2018) Erratum: JAK inhibition as a therapeutic
strategy for immune and inflammatory diseases. Nat Rev Drug Discov 17:78. https://doi.org/
10.1038/nrd.2017.267
10. Chen C, Yin Y, Shi G et al (2022) A highly selective JAK3 inhibitor is developed for
treating rheumatoid arthritis by suppressing γc cytokine–related JAK-STAT signal. Sci Adv
8:eabo4363. https://doi.org/10.1126/sciadv.abo4363
11. Sadybekov AA, Sadybekov AV, Liu Y et al (2022) Synthon-based ligand discovery in virtual
libraries of over 11 billion compounds. Nature 601:452–459. https://doi.org/10.1038/s41586-
021-04220-9
12. Lionta E, Spyrou G, Vassilatis KD, Cournia Z (2014) Structure-based virtual screening for
drug discovery: principles, applications and recent advances. Curr Top Med Chem 14:1923–
1938
13. Kar S, Roy K (2013) How far can virtual screening take us in drug discovery? Expert Opin
Drug Discov 8:245–261. https://doi.org/10.1517/17460441.2013.761204
14. Bissantz C, Kuhn B, Stahl M (2010) A medicinal chemist’s guide to molecular interactions.
J Med Chem 53:5061–5084. https://doi.org/10.1021/jm100112j
15. Burley SK (2021) Impact of structural biologists and the Protein Data Bank on small-molecule
drug discovery and development. J Biol Chem. https://doi.org/10.1016/j.jbc.2021.100559
16. Chakraborti S, Chakraborty M, Bose A et al (2021) Identification of potential binders of MTB
Universal Stress Protein (Rv1636) through an in silico approach and insights into compound
selection for experimental validation. Front Mol Biosci 8:599221. https://doi.org/10.3389/
fmolb.2021.599221
17. Verma H, Khatri B, Chakraborti S, Chatterjee J (2018) Increasing the bioactive space of
peptide macrocycles by thioamide substitution. Chem Sci 9:2443–2451. https://doi.org/10.
1039/C7SC04671E
18. Capra JA, Singh M (2007) Predicting functionally important residues from sequence
conservation. Bioinformatics 23:1875–1882. https://doi.org/10.1093/bioinformatics/btm270
19. Berman HM, Westbrook J, Feng Z et al (2000) The Protein Data Bank. Nucleic Acids Res
28:235–242. https://doi.org/10.1093/nar/28.1.235
20. Rhodes G (2006) An overview of protein crystallography. In: Rhodes GBT-CMCC (ed)
Complementary science. Academic Press, Burlington, pp 7–30
21. Howard MJ (1998) Protein NMR spectroscopy. Curr Biol 8:R331–R333. https://doi.org/10.
1016/S0960-9822(98)70214-3
22. Savva C (2019) A beginner’s guide to cryogenic electron microscopy. Biochem (Lond) 41:46–
52. https://doi.org/10.1042/BIO04102046
23. Webb B, Eswar N, Fan H, Khuri N, Pieper U, Dong GQ, Sali A (2014) Comparative modeling
of drug target proteins. In: Reedijk J (ed) Elsevier reference module in chemistry, molecular
sciences and chemical engineering. Elsevier, Waltham. https://doi.org/10.1016/B978-0-12-
409547-2.11133-3
24. Varadi M, Anyango S, Deshpande M et al. (2022) AlphaFold protein structure database:
massively expanding the structural coverage of protein-sequence space with high-accuracy
models. Nucleic Acids Res 50(D1): D439–D444. https://doi.org/10.1093/nar/gkab1061.
PMID: 34791371; PMCID: PMC8728224
25. Bowie JU, Lüthy R, Eisenberg D (1991) A method to identify protein sequences that fold into
a known three-dimensional structure. Science 253:164–170. https://doi.org/10.1126/science.
1853201
26. Zhou H, Skolnick J (2013) FINDSITEcomb: a threading/structure-based, proteomic-scale
virtual ligand screening approach. J Chem Inf Model 53:230–240. https://doi.org/10.1021/
ci300510n
27. Teilum K, Olsen JG, Kragelund BB (2009) Functional aspects of protein flexibility. Cell Mol
Life Sci 66:2231–2247. https://doi.org/10.1007/s00018-009-0014-6
20 S. Chakraborti and S. Sachchidanand

28. Wlodawer A, Minor W, Dauter Z, Jaskolski M (2008) Protein crystallography for non-
crystallographers, or how to get the best (but not more) from published macromolecular
structures. FEBS J 275:1–21. https://doi.org/10.1111/j.1742-4658.2007.06178.x
29. Rupp B (2009) Biomolecular crystallography: principles, practice, and application to
structural biology, 1st ed. Garland Science
30. Pozharski E, Weichenberger CX, Rupp B (2013) Techniques, tools and best practices for ligand
electron-density analysis and results from their application to deposited crystal structures. Acta
Crystallogr D Biol Crystallogr 69:150–167. https://doi.org/10.1107/S0907444912044423
31. Davis AM, St-Gallay SA, Kleywegt GJ (2008) Limitations and lessons in the use of X-ray
structural information in drug design. Drug Discov Today 13:831–841. https://doi.org/10.
1016/j.drudis.2008.06.006
32. Hu Y, Cheng K, He L et al (2021) NMR-based methods for protein analysis. Anal Chem
93:1866–1879. https://doi.org/10.1021/acs.analchem.0c03830
33. Sekhar A, Kay LE (2013) NMR paves the way for atomic level descriptions of sparsely popu-
lated, transiently formed biomolecular conformers. Proc Natl Acad Sci 110:12867–12874.
https://doi.org/10.1073/pnas.1305688110
34. Van Drie JH, Tong L (2020) Cryo-EM as a powerful tool for drug discovery. Bioorg Med
Chem Lett 30:127524. https://doi.org/10.1016/j.bmcl.2020.127524
35. Subramaniam S, Earl LA, Falconieri V et al (2016) Resolution advances in cryo-EM enable
application to drug discovery. Curr Opin Struct Biol 41:194–202. https://doi.org/10.1016/j.
sbi.2016.07.009
36. Cavasotto CN, Palomba D (2015) Expanding the horizons of G protein-coupled receptor
structure-based ligand discovery and optimization using homology models. Chem Commun
51:13576–13594. https://doi.org/10.1039/C5CC05050B
37. Jumper J, Evans R, Pritzel A et al (2021) Highly accurate protein structure prediction with
AlphaFold. Nature. https://doi.org/10.1038/s41586-021-03819-2
38. Baek M, DiMaio F, Anishchenko I et al (2021) Accurate prediction of protein structures
and interactions using a three-track neural network. Science 373:871–876. https://doi.org/10.
1126/science.abj8754
39. Lee C, Su B-H, Tseng YJ (2022) Comparative studies of AlphaFold, RoseTTAFold and
modeller: a case study involving the use of G-protein-coupled receptors. Brief Bioinform
bbac308. https://doi.org/10.1093/bib/bbac308
40. Heo L, Arbour CF, Feig M (2019) Driven to near-experimental accuracy by refinement via
molecular dynamics simulations. Proteins Struct Funct Bioinforma 87:1263–1275. https://
doi.org/10.1002/prot.25759
41. Zhang Y, Vass M, Shi D et al (2022) Benchmarking refined and unrefined AlphaFold2
structures for hit discovery. ChemRxiv. https://doi.org/10.26434/chemrxiv-2022-kcn0d-v2
42. Schiebel J, Krimmer SG, Röwer K et al. (2016) High-throughput crystallography: reliable
and efficient identification of fragment hits. Structure 24(8): 1398–1409. ISSN 0969-2126,
https://doi.org/10.1016/j.str.2016.06.010
43. Wu B, Barile E, De SK, Wei J, Purves A, Pellecchia M (2015) High-throughput screening by
nuclear magnetic resonance (HTS by NMR) for the identification of PPIs antagonists. Curr
Top Med Chem 15(20):2032–2042. https://doi.org/10.2174/1568026615666150519102459
44. Gorgulla C, Boeszoermenyi A, Wang ZF et al. (2020) An open-source drug discovery platform
enables ultra-large virtual screens. Nature 580: 663–668. https://doi.org/10.1038/s41586-020-
2117-z
45. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ (1997) Experimental and computational
approaches to estimate solubility and permeability in drug discovery and development settings.
Adv Drug Deliv Rev 23:3–25. https://doi.org/10.1016/s0169-409x(00)00129-0
46. Baell JB, Holloway GA (2010) New substructure filters for removal of pan assay interference
compounds (PAINS) from screening libraries and for their exclusion in bioassays. J Med
Chem 53:2719–2740. https://doi.org/10.1021/jm901137j
47. Walters WP, Stahl MT, Murcko MA (1998) Virtual screening—an overview. Drug Discov
Today 3:160–178. https://doi.org/10.1016/S1359-6446(97)01163-X
1 SBDD and Its Challenges 21

48. Erlanson DA, Fesik SW, Hubbard RE et al (2016) Twenty years on: the impact of fragments
on drug discovery. Nat Rev Drug Discov 15:605–619. https://doi.org/10.1038/nrd.2016.109
49. John Harris C, Hill R, Sheppard D et al (2011) The design and application of target-focused
compound libraries. Comb Chem High Throughput Screen 14:521–531
50. Moret N, Clark NA, Hafner M et al (2019) Cheminformatics tools for analyzing and designing
optimized small-molecule collections and libraries. Cell Chem Biol 26:765-777.e3. https://
doi.org/10.1016/j.chembiol.2019.02.018
51. Masimirembwa CM, Bredberg U, Andersson TB (2003) Metabolic stability for drug discovery
and development. Clin Pharmacokinet 42:515–528. https://doi.org/10.2165/00003088-200
342060-00002
52. Schnider P (2021) Overview of strategies for solving ADMET challenges. In: The medicinal
chemist’s guide to solving ADMET challenges. Royal Society of Chemistry, pp 1–15
53. Kumar S, Sharma R, Roychowdhury A (2012) Modulation of cytochrome-P450 inhibition
(CYP) in drug discovery: a medicinal chemistry perspective. Curr Med Chem 19:3605–3621
54. Spyrakis F, Cozzini P, Kellogg GE (2010) Docking and scoring in drug discovery. Burger’s
Med Chem Drug Discov 601–684
55. Bitencourt-Ferreira G, de Azevedo WF (2019) Machine learning to predict binding affinity
BT. In: de Azevedo Jr. WF (ed) Docking screens for drug discovery. Springer, New York, pp
251–273
56. Jones D, Kim H, Zhang X et al (2021) Improved protein-ligand binding affinity prediction
with structure-based deep fusion inference. J Chem Inf Model 61:1583–1592. https://doi.org/
10.1021/acs.jcim.0c01306
57. Thafar M, Bin RA, Albaradei S et al (2019) Comparison study of computational predic-
tion tools for drug-target binding affinities. Front Chem. https://doi.org/10.3389/fchem.2019.
00782
58. Dhakal A, McKay C, Tanner JJ, Cheng J (2022) Artificial intelligence in the predic-
tion of protein–ligand interactions: recent advances and future directions. Brief Bioinform
23:bbab476. https://doi.org/10.1093/bib/bbab476
59. Dutta S, Bose K (2021) Remodelling structure-based drug design using machine learning.
Emerg Top Life Sci 5:13–27. https://doi.org/10.1042/ETLS20200253
60. Chakraborti S, Hatti K, Srinivasan N (2021) ‘All that glitters is not gold’: high-resolution
crystal structures of ligand-protein complexes need not always represent confident binding
poses. Int J Mol Sci. https://doi.org/10.3390/ijms22136830
61. Blow DM (2002) Rearrangement of Cruickshank’s formulae for the diffraction-component
precision index. Acta Crystallogr Sect D 58:792–797. https://doi.org/10.1107/S09074449020
03931
62. Cruickshank DWJ (1999) Remarks about protein structure precision. Acta Crystallogr Sect
D 55:583–601. https://doi.org/10.1107/S0907444998012645
63. Deller MC, Rupp B (2015) Models of protein-ligand crystal structures: trust, but verify. J
Comput Aided Mol Des 29:817–836. https://doi.org/10.1007/s10822-015-9833-8
64. Tickle IJ (2012) Statistical quality indicators for electron-density maps. Acta Crystallogr Sect
D 68:454–467. https://doi.org/10.1107/S0907444911035918
65. Meyder A, Nittinger E, Lange G et al (2017) Estimating electron density support for individual
atoms and molecular fragments in x-ray structures. J Chem Inf Model 57:2437–2447. https:/
/doi.org/10.1021/acs.jcim.7b00391
66. Joosten RP, Joosten K, Murshudov GN, Perrakis A (2012) PDB_REDO: constructive vali-
dation, more than just looking for errors. Acta Crystallogr D Biol Crystallogr 68:484–496.
https://doi.org/10.1107/S0907444911054515
67. Nishi H, Tyagi M, Teng S et al (2013) Cancer missense mutations alter binding properties of
proteins and their interaction networks. PLoS ONE 8:e66273
68. Konc J, Česnik T, Konc JT et al (2012) ProBiS-database: precalculated binding site similarities
and local pairwise alignments of PDB structures. J Chem Inf Model 52:604–612. https://doi.
org/10.1021/ci2005687
22 S. Chakraborti and S. Sachchidanand

69. Le Guilloux V, Schmidtke P, Tuffery P (2009) Fpocket: an open source platform for ligand
pocket detection. BMC Bioinform 10:168. https://doi.org/10.1186/1471-2105-10-168
70. Halgren TA (2009) Identifying and characterizing binding sites and assessing druggability. J
Chem Inf Model 49:377–389. https://doi.org/10.1021/ci800324m
71. McGovern SL, Shoichet BK (2003) Information decay in molecular docking screens against
Holo, Apo, and modeled conformations of enzymes. J Med Chem 46:2895–2907. https://doi.
org/10.1021/jm0300330
72. Koshland DE (1958) Application of a theory of enzyme specificity to protein synthesis*. Proc
Natl Acad Sci 44:98–104. https://doi.org/10.1073/pnas.44.2.98
73. Kim K-H, Kim ND, Seong B-L (2010) Pharmacophore-based virtual screening: a review of
recent applications. Expert Opin Drug Discov 5:205–222. https://doi.org/10.1517/174604410
03592072
74. Sherman W, Day T, Jacobson MP et al (2006) Novel procedure for modeling ligand/receptor
induced fit effects. J Med Chem 49:534–553. https://doi.org/10.1021/jm050540c
75. Hollingsworth SA, Dror RO (2018) Molecular dynamics simulation for all. Neuron 99:1129–
1143. https://doi.org/10.1016/j.neuron.2018.08.011
76. Webb B, Sali A (2016) Comparative protein structure modeling using MODELLER. Curr
Protoc Bioinforma 54:5.6.1–5.6.37. https://doi.org/10.1002/cpbi.3
77. Jacobson MP, Pincus DL, Rapp CS et al (2004) A hierarchical approach to all-atom protein
loop prediction. Proteins Struct Funct Bioinforma 55:351–367. https://doi.org/10.1002/prot.
10613
78. Schmid S, Hugel T (2020) Controlling protein function by fine-tuning conformational
flexibility. Elife 9:e57180. https://doi.org/10.7554/eLife.57180
79. Möbitz H (2015) The ABC of protein kinase conformations. Biochim Biophys Acta - Proteins
Proteomics 1854:1555–1566. https://doi.org/10.1016/j.bbapap.2015.03.009
80. Huse M, Kuriyan J (2002) The conformational plasticity of protein kinases. Cell 109:275–282.
https://doi.org/10.1016/S0092-8674(02)00741-9
81. Wang X, Kim J (2012) Conformation-specific effects of Raf kinase inhibitors. J Med Chem
55:7332–7341. https://doi.org/10.1021/jm300613w
82. Tong M, Seeliger MA (2015) Targeting conformational plasticity of protein kinases. ACS
Chem Biol 10:190–200. https://doi.org/10.1021/cb500870a
83. Kwarcinski FE, Brandvold KR, Phadke S et al (2016) Conformation-selective analogues
of dasatinib reveal insight into kinase inhibitor binding and selectivity. ACS Chem Biol
11:1296–1304. https://doi.org/10.1021/acschembio.5b01018
84. Su M-G, Weng JT-Y, Hsu JB-K et al (2017) Investigation and identification of functional post-
translational modification sites associated with drug binding and protein-protein interactions.
BMC Syst Biol 11:132. https://doi.org/10.1186/s12918-017-0506-1
85. Liwo A, Czaplewski C, Ołdziej S, Scheraga HA (2008) Computational techniques for efficient
conformational sampling of proteins. Curr Opin Struct Biol 18:134–139. https://doi.org/10.
1016/j.sbi.2007.12.001
86. Haddad Y, Adam V, Heger Z (2020) Ten quick tips for homology modeling of high-resolution
protein 3D structures. PLOS Comput Biol 16:e1007449
87. Hameduh T, Haddad Y, Adam V, Heger Z (2020) Homology modeling in the time of collective
and artificial intelligence. Comput Struct Biotechnol J 18:3494–3506. https://doi.org/10.1016/
j.csbj.2020.11.007
88. Sastry MG, Adzhigirey M, Day T et al (2013) Protein and ligand preparation: parameters,
protocols, and influence on virtual screening enrichments. J Comput Aided Mol Des 27:221–
234. https://doi.org/10.1007/s10822-013-9644-8
89. Vriend G (1990) WHAT IF: a molecular modeling and drug design program. J Mol Graph
8:52–56. https://doi.org/10.1016/0263-7855(90)80070-V
90. Kim S, Chen J, Cheng T et al (2018) PubChem 2019 update: improved access to chemical
data. Nucleic Acids Res 47:D1102–D1109. https://doi.org/10.1093/nar/gky1033
91. Mendez D, Gaulton A, Bento AP et al (2018) ChEMBL: towards direct deposition of bioassay
data. Nucleic Acids Res 47:D930–D940. https://doi.org/10.1093/nar/gky1075
1 SBDD and Its Challenges 23

92. Gilson MK, Liu T, Baitaluk M et al (2016) BindingDB in 2015: A public database for
medicinal chemistry, computational chemistry and systems pharmacology. Nucleic Acids
Res 44:D1045–D1053. https://doi.org/10.1093/nar/gkv1072
93. Cousins KR (2005) ChemDraw Ultra 9.0. CambridgeSoft, 100 CambridgePark Drive,
Cambridge, MA 02140. www.cambridgesoft.com. See Web site for pricing options. J Am
Chem Soc 127:4115–4116. https://doi.org/10.1021/ja0410237
94. Cole DJ, Horton JT, Nelson L, Kurdekar V (2019) The future of force fields in computer-aided
drug design. Future Med Chem 11:2359–2363. https://doi.org/10.4155/fmc-2019-0196
95. O’Boyle NM, Banck M, James CA et al (2011) Open Babel: an open chemical toolbox. J
Cheminform 3:33. https://doi.org/10.1186/1758-2946-3-33
96. Gohlke H, Klebe G (2002) Approaches to the description and prediction of the binding affinity
of small-molecule ligands to macromolecular receptors. Angew Chemie Int Ed 41:2644–2676.
https://doi.org/10.1002/1521-3773(20020802)41:15<2644::AID-ANIE2644>3.0.CO;2-O
97. Pantsar T, Poso A (2018) Binding affinity via docking: fact and fiction. Molecules 23:1899.
https://doi.org/10.3390/molecules23081899
98. Plewczynski D, Łaźniewski M, Augustyniak R, Ginalski K (2011) Can we trust docking
results? Evaluation of seven commonly used programs on PDBbind database. J Comput
Chem 32:742–755. https://doi.org/10.1002/jcc.21643
99. van Gunsteren WF, Daura X, Fuchs PFJ et al (2021) On the effect of the various assumptions
and approximations used in molecular simulations on the properties of bio-molecular systems:
overview and perspective on issues. ChemPhysChem 22:264–282. https://doi.org/10.1002/
cphc.202000968
100. Anighoro A (2020) Underappreciated chemical interactions in protein–ligand complexes
BT—quantum mechanics in drug discovery. In: Heifetz A (ed). Springer US, New York,
pp 75–86
101. Zhang X, Gong Z, Li J, Lu T (2015) Intermolecular sulfur···oxygen interactions: theoretical
and statistical investigations. J Chem Inf Model 55:2138–2153. https://doi.org/10.1021/acs.
jcim.5b00177
102. Ferreira LG, Dos Santos RN, Oliva G, Andricopulo AD (2015) Molecular docking and
structure-based drug design strategies. Moleculee 20:13384–13421
103. Fan M, Wang J, Jiang H et al (2021) GPU-accelerated flexible molecular docking. J Phys
Chem B 125:1049–1060. https://doi.org/10.1021/acs.jpcb.0c09051
104. Wang L, Chambers J, Abel R (2019) Protein-ligand binding free energy calculations with
FEP+ BT. In: Bonomi M, Camilloni C (eds) Biomolecular simulations: methods and protocols.
Springer, New York, pp 201–232
105. van der Kamp MW, Mulholland AJ (2013) Combined quantum mechanics/molecular
mechanics (QM/MM) methods in computational enzymology. Biochemistry 52:2708–2728.
https://doi.org/10.1021/bi400215w
106. Cao L, Ryde U (2018) On the difference between additive and subtractive QM/MM
calculations. Front Chem. https://doi.org/10.3389/fchem.2018.00089
107. Ladbury JE (1996) Just add water! The effect of water on the specificity of protein-ligand
binding sites and its potential application to drug design. Chem Biol 3:973–980. https://doi.
org/10.1016/S1074-5521(96)90164-7
108. Zsidó BZ, Hetényi C (2021) The role of water in ligand binding. Curr Opin Struct Biol 67:1–8.
https://doi.org/10.1016/j.sbi.2020.08.002
109. Klebe G (2011) On the validity of popular assumptions in computational drug design. J
Cheminform 3:O18. https://doi.org/10.1186/1758-2946-3-S1-O18
110. Yang Y, Lightstone FC, Wong SE (2013) Approaches to efficiently estimate solvation and
explicit water energetics in ligand binding: the use of WaterMap. Expert Opin Drug Discov
8:277–287. https://doi.org/10.1517/17460441.2013.749853
111. Cappel D, Sherman W, Beuming T (2017) Calculating water thermodynamics in the binding
site of proteins—applications of WaterMap to drug discovery. Curr Top Med Chem 17:2586–
2598
24 S. Chakraborti and S. Sachchidanand

112. Lyne PD, Lamb ML, Saeh JC (2006) Accurate prediction of the relative potencies of members
of a series of kinase inhibitors using molecular docking and MM-GBSA scoring. J Med Chem
49:4805–4808. https://doi.org/10.1021/jm060522a
113. Spyrakis F, BidonChanal A, Barril X, Javier Luque F (2011) Protein flexibility and ligand
recognition: challenges for molecular modeling. Curr Top Med Chem 11:192–210
114. Gaudreault F, Chartier M, Najmanovich R (2012) Side-chain rotamer changes upon
ligand binding: common, crucial, correlate with entropy and rearrange hydrogen bonding.
Bioinformatics 28:i423–i430. https://doi.org/10.1093/bioinformatics/bts395
115. Alvarez-Garcia D, Barril X (2014) Relationship between protein flexibility and binding:
lessons for structure-based drug design. J Chem Theory Comput 10:2608–2614. https://doi.
org/10.1021/ct500182z
116. Lin X (2022) Applications of molecular dynamics simulations in drug discovery. In: Tripathi
T, Dubey VK (eds) Advances in protein molecular and structural biology methods. Academic
Press, pp 455–465
117. Amaro RE, Baudry J, Chodera J et al (2018) Ensemble docking in drug discovery. Biophys J
114:2271–2278
118. Tian S, Sun H, Pan P et al (2014) Assessing an ensemble docking-based virtual screening
strategy for kinase targets by considering protein flexibility. J Chem Inf Model 54:2664–2679.
https://doi.org/10.1021/ci500414b
119. Shaw DE, Dror RO, Salmon JK et al (2009) Millisecond-scale molecular dynamics simulations
on anton. In: Proceedings of the conference on high performance computing networking,
storage and analysis. Association for Computing Machinery, New York
120. Ngo VA, Garcia AE (2022) Millisecond molecular dynamics simulations of KRas-dimer
formation and interfaces. Biophys J. https://doi.org/10.1016/j.bpj.2022.04.026
121. Pandey M, Fernandez M, Gentile F et al (2022) The transformational role of GPU computing
and deep learning in drug discovery. Nat Mach Intell 4:211–221. https://doi.org/10.1038/s42
256-022-00463-x
122. Puertas-Martín S, Banegas-Luna AJ, Paredes-Ramos M et al (2020) Is high performance
computing a requirement for novel drug discovery and how will this impact academic efforts?
Expert Opin Drug Discov 15:981–985. https://doi.org/10.1080/17460441.2020.1758664
Chapter 2
In Silico Discovery of Class IIb HDAC
Inhibitors: The State of Art

Samima Khatun, Sk. Abdul Amin, Shovanlal Gayen, and Tarun Jha

Abstract HDAC6 and HDAC10 are class IIb HDAC isoenzymes. They have unique
structural and physiological functions. They are key regulators of different physio-
logical and pathological disease conditions. HDAC6 and HDAC10 are involved in
different signaling pathways associated with several neurological disorders, various
cancers at early as well as advanced stages, rare diseases, immunological conditions,
etc. Thus, targeting these two enzymes has been found to be effective for various ther-
apeutic purposes in recent years. More work is still needed to pinpoint the selectivity
as well as potency of class IIb HDAC inhibitors (HDACi) for their clinical devel-
opment. The present chapter deals with the structural biology of class IIb HDACs
and discusses how in silico studies including the virtual screening approaches have
been implemented to design HDAC6 and HDAC10 inhibitors. In addition, the inter-
actions of class IIb HDACs with their inhibitors are also highlighted extensively to
get a detail insight. This chapter offers understanding for designing newer class IIb
HDAC inhibitors in future.

Keywords HDAC6 · HDAC10 · Drug design and discovery · QSAR · Molecular

docking · MD simulation

2.1 Introduction

Epigenetic alterations caused by genetic flaws result in functional dysregulation of

epigenetic regulators or proteins [1–4]. It eventually leads to changes in protein
expression, which play a significant role in a variety of human diseases, including

S. Khatun · S. Gayen
Laboratory of Drug Design and Discovery, Department of Pharmaceutical Technology, Jadavpur
University, Kolkata 700032, India
Sk. A. Amin · T. Jha (B)
Natural Science Laboratory, Division of Medicinal and Pharmaceutical Chemistry, Department of
Pharmaceutical Technology, Jadavpur University, Kolkata 700032, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 25

S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_2
26 S. Khatun et al.

various types of cancer, cardiovascular diseases, infections, inflammatory diseases,

and neurological disorders [5–7]. A better understanding and application of epige-
netics will aid in the identification of novel therapeutic treatments in the form of
personalized medicine for various diseases [5]. Several histone post-translational
modifications are critical epigenetic modulators and they are believed to influence
gene expression [8]. Several post-translational modifications on histone proteins
include: (1) the acetylation of specific lysine residues (by histone acetyltransferases),
(2) the methylation of lysine and arginine residues (by histone methyltransferases),
and (3) the phosphorylation of specific serine groups (by histone kinases) [9]. The
amino termini of the histone proteins are utilized by the transcription regulators to
carry out a number of post-translational modifications [10, 11].
Among the different post-translational modifications, the most researched process
is the acetylation and deacetylation of histones. It takes place at the lysine amino
termini and is controlled by both histone acetyltransferases (HATs) and histone
deacetylases (HDACs) [12–14]. Due to the imbalance between HAT and HDAC,
dysregulation of genetic expression results in chromatin instability and epigenetic
diseases or disorders (Fig. 2.1). While HDAC inhibition results in continuous expres-
sion of the targeted gene, HAT inhibition results in the inexpression of the targeted
gene [15–19]. It is well known that the overexpression of HDACs contributes to a
variety of cancers as well as other neurological, autoimmune, inflammatory, cardiac,
and pulmonary diseases [20–22]. HDACs are also found to deacetylate a number of
non-histone proteins, including p53, E2F, α-tubulin, and Myo D. This leads to much
more complex roles for HDACs in numerous other cellular processes. As a result,
HDAC inhibition has drawn considerable interest and grown in importance as a drug
target.

Fig. 2.1 Histone modification by HAT and HDAC

2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 27

There are 18 isoforms of mammalian HDACs recognized until now. Based on

their similarity with yeast protein and method of action, they are classified into four
separate types [23–25]. The structure, enzymatic function, sub-cellular localization,
and expression patterns of the four classes are unique [26]. The HDAC isozymes were
numbered in the order in which they were discovered. HDAC1, HDAC2, HDAC3,
and HDAC8 from class I have a sequence similarity to yeast- reduced potassium
dependence (Rpd3)-like proteins [27–30]. They are mostly found in the nucleus
and work primarily through histone proteins as their substrate. Class II HDACs
have a similar amino acid sequence to yeast histone deacetylase 1 (Hda1). Based
on sequence homology and domain structure, they were further categorized into
two sub-classes, IIa (HDAC 4, 5, 7, and 9) and IIb (HDAC 6 and 10) [31–33].
Class IIa HDACs are located in the nucleus and they shuttle to the cytoplasm after
being phosphorylated by kinases. Class IIb HDACs, on the other hand, are known
to be cytoplasmic and act through diverse non-histone proteins as their substrates
for deacetylase activity. Table 2.1 highlights the classification of class IIb HDAC
isoforms as well as their cellular location and physiological roles. Class III HDACs
or sirtuins, which include SIRT1, 2, 3, 4, 5, and 6, were so termed because they
resemble the proteins that silence the yeast Sir2 gene [34]. Only HDAC11, which
is known to have similarities to both class I and II catalytic domains of HDACs,
belongs to class IV HDAC [35]. Class I, II, and IV HDACs are in the family of
zinc-dependent HDAC isoforms, where the cofactor for the hydrolysis of acetylated
substrates is the metal ion Zn2+ . Aside from that, class III HDACs depend on NAD+,
which serves as a cofactor for their enzymatic activity [25].
A zinc-binding group (ZBG) that interacts with the zinc ion at the catalytic pocket,
a cap group that interacts with the surface of the enzyme, and a linker that acts as a
link between the cap and ZBG make up the canonical feature of HDAC inhibitors
(HDACi) [36, 37]. Belinostat, panobinostat, romidepsin, and vorinostat (SAHA) have
all received clinical approval for the treatment of lymphoma and multiple myeloma

Table 2.1 Classification of HDAC class IIb isoform, their cellular localization, and functions
Class HDAC Chromosomal Amino Cellular Location in Physiological
isoform location acids no localization body/ function
expression
IIb HDAC6 Xp11.23 1215 Cytoplasm Tissue Regulation of
specific protein degradation
through aggresome
pathway, Hsp90
chaperone activity,
cytoskeletal
dynamics, cell
motility,
angiogenesis
HDAC10 2q13.33 669 Angiogenesis,
autophagy,
neurodegeneration
28 S. Khatun et al.

[38]. Due to their non-selectivity and broad-spectrum activity, these approved non-
selective pan-HDACis are said to have a number of side effects, including exhaustion,
nausea/vomiting, cardiotoxicity, etc. Therefore, there is a growing need for isoform-
specific HDACi in order to reduce the side effects as well as apply them in more
focused and selective treatments of a specific disease condition.
Until now, many HDAC isoforms have been investigated and their inhibitors are
thoroughly defined. Since its first discovery in 1999, HDAC6, a class IIB HDAC
isoform, has attracted also important attention among the different HDACs [39].
HDAC6 is a physically and functionally distinct cytoplasmic deacetylase. It is known
for its deacetylase activity of certain cytosolic non-histone substrates like heat shock
protein (Hsp90), cortactin, peroxiredoxin, α-tubulin, heat shock transcription facto-
1 (HSF-1), etc. [40]. The first identified and extensively researched physiological
substrate of HDAC6 is α-tubulin. HDAC6 controls the acetylation of lysine 40
in α-tubulin [41]. It is also known to participate in the tumorigenesis along with
the development and metastasis through various pathways such as tubulin, Hsp90,
and protein ubiquitination [42]. The effectiveness of selective HDAC6 inhibition
in treating cancers such as bladder cancer, malignant melanoma, and lung cancer
as well as neurodegenerative diseases such as Alzheimer’s disease, Huntington’s
disease, and Parkinson’s disease has also been extensively demonstrated in some
studies and reports [43, 44]. The application of HDAC6i in rare disorders such as
Rett syndrome, Charcot–Marie–Tooth disease, and amyotrophic lateral sclerosis has
also been shown in recent investigations [45]. According to research using HDAC6
mutant mice, selective HDAC6 inhibitors are less cytotoxic to normal cells than pan-
HDAC inhibitors, which mitigate their negative effects [46]. Tubacin, Tubastatin A,
ACY-1215, ACY-241, and Nexturastat A are examples of specific HDAC6is. Recent
studies on the X-ray crystal structures of HDAC6 CD2 and HDAC6i complexes have
shed light on the structure and catalytic mechanism of the molecular characteristics
that determines binding affinity for the target [46].
HDAC10 is also an important member in HDACs family and it shares struc-
tural similarities with HDAC6. The HDAC10 gene is located on chromosome 22
[47]. It is made up of 20 exons and two spliced transcripts. It comprises an N-
terminal catalytic domain and a C-terminal leucine-rich domain. The expression of
HDAC10, a member of the arginase/deacetylase superfamily, varies between the cyto-
plasm and the nucleus. Additionally, HDAC10 is expressed in the majority of human
tissues, including the heart, liver, spleen, pancreas, placenta, kidney, and testicles.
Recent research has shown that HDAC10 controls polyamine levels and functions
as a polyamine deacetylase (PDAC) [47]. The pathogenesis of many malignancies is
thought to include histone deacetylase 10 (HDAC10), and pharmacological blockade
of this enzyme might aid in reversing the malignant phenotypes. A number of studies
have highlighted HDAC10 inhibitors as possible anti-cancer agent [48–51]. However,
the absence of the crystal structure of human HDAC10 hinders the structure-based
rational drug design effort. HDAC10, the only other HDAC class IIb member, has
received minimal attention from the medicinal chemistry community.
Prior to the 1960s, traditional drug discovery efforts in the pharmaceutical industry
were largely committed to assess natural and synthetic chemicals against a specific
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 29

biological endpoint [53–55]. After a potential drug or drug-like substance has been
narrowed down from thousands of natural and synthetic compounds through an
arduous process, medicinal chemists would then synthesize hundreds of related
compounds (derivatives or analogs) to determine which molecule is both the safest
and the most potent [56–60]. As a result, the costs and possible dangers with this
approach dramatically increased.
The conventional drug design process underwent a significant change in the 1960s.
The paradigms were shifted by a number of major publications released in the 1960s
[53]. Since then, rational drug design (RDD) paradigms have attracted more atten-
tion for the design of new chemical entities as well as the optimization of chemical
structures for improved biological activity. The development of computational chem-
istry, protein crystallography, and molecular biology in the early 1980s considerably
benefited RDD paradigms in their efforts to improve the accuracy of binding affinity
predictions. With fast advancing high throughput screening (HTS) and combina-
torial chemical technology, computer-aided drug design (CADD) methodologies,
and effective contributions, rational drug discovery has now become increasingly
interdisciplinary [61].
In silico work such as quantitative structure–activity relationship (QSAR), phar-
macophore mapping, virtual screening, homology modeling, molecular docking and
molecular dynamic (MD) simulation studies applied to design class IIb HDAC
inhibitors is the main emphasis of this chapter. The present chapter is broken into
two sections. The first section presents an overview of class IIb HDACs (HDAC6 and
HDAC10) as well as their structural biology, functions, and mechanism of action. The
next section discusses about the different approaches related to in silico discovery
of class IIb HDAC inhibitors. This chapter may be useful in the future for designing
highly active as well as selective HDAC class IIb inhibitors.

2.2 Structural Biology of HDAC6

Among the different HDACs, HDAC6 has distinctive structural and functional char-
acteristic. Its localization is mainly in the cytoplasm and primarily responsible for
deacetylation of non-histone proteins like α-tubulin of microtubule. Unlike other
HDACs, HDAC6 has two catalytic deacetylase domains in its structure as shown in
Fig. 2.2. It has also an ubiquitin binding domain that is responsible to form aggre-
somes for degradation of polyubiquitinated misfolded proteins [62]. The N-terminal
domain of HDAC6 is comprised two domains, namely nuclear localized signal (NLS)
and nuclear export signal (NES). NLS is rich in arginine and lysine amino acids, while
NES is rich in leucine amino acid. The two catalytic domains 1 and 2 are comprised
amino acids 88–447 and 482–800, respectively. The dynein motor binding region
(DMB) is connecting the two catalytic domains. The SE14 domain which is a Ser-
Glu tetrapeptide domain sequence (SE14) is important for intracellular retention and
tau interaction of HDAC6.
30 S. Khatun et al.

Nuclear Localization Dynein motor zinc-finger ubiquitin-

signal binding binding domain

NLS NES DMB SE14 NES

Catalytic Domain 1 Catalytic Domain 2 ZnF

Nuclear export serine-glutamate

signal tetradecapeptide repeat

Fig. 2.2 Different domains in HDAC6

Hubbert et al. [41] first reported the in vivo and in vitro tubulin deacetylation
for HDAC6. From the study, they have proved that HDAC6 is responsible for
microtubule-dependent cell motility. Among the two catalytic domains, only one
catalytic domain binds to the tubulin and is important for deacetylation of tubulin
[63]. It is documented that CD2 of HDAC6 has tubulin deacetylation function and
the function of CD1 is not revealed [64]. Several post-translational modifications
including acetylation [65], sumoylation, ubiquitination [66], phosphorylation, etc.
are responsible for the regulation of deacetylase activity of HDAC6.

2.2.1 Insight into HDAC6 Crystal Structures

Several X-ray crystal structures of HDAC6 were reported from Homo sapiens
(human) and Danio rerio (zebrafish), and the structures give important insights in
ligand receptor interactions for different HDAC6 inhibitors. The crystal structures
of both the catalytic domains have been studied extensively and reported [67]. The
crystal structures of both catalytic domains from zebrafish HDAC6 in complex with
inhibitors were also reported [68].
Figure 2.3 shows the important amino acids responsible for ligand–receptor inter-
actions for HDAC6 with enantiomers of trichostatin A (TSA). The isolated catalytic
domain 1 contains (R)-TSA in its catalytic center (Fig. 2.3a). The crystal structure
of catalytic domain 2 with (S)-TSA bound is shown in Fig. 2.3b. The analysis of the
two structures revealed that the backbone structures of both catalytic domains are
very similar. The ligand binding sites in the two structures are highly conserved and
clearly point out the importance of the narrow hydrophobic channel in the binding
site. P83, F202, W261, etc. are important residues in case of catalytic domain 1,
and F643, L712, etc. are important in catalytic domain 2. Several important residues
are found in both the structures for interaction with the Zn2+ ions. An important
difference found between the two structures is the presence of bulkier amino acid
W261 in catalytic domain 1, whereas in catalytic domain 2, the amino acid is F643.
Thus, (S)-TSA can selectively bind to HDAC6 over other HDACs in that its cap
group interacts with F463. It has been also proved that catalytic domain of HDAC6
can accommodate different substrates, whereas catalytic domain 1 is highly specific
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 31

Fig. 2.3 Ligand–receptor interactions for HDAC6 with (R) and (S)-enantiomers of trichostatin A
(TSA)

for the hydrolysis of C-terminal of acetyl-lysine residues. In different HDACs, the

inhibitor is involved in bidentate coordination with the Zn2+ ion in its ligand–receptor
interactions, whereas a specific inhibitor of HDAC6 is involved in monodentate coor-
dination with the Zn2+ revealing unique enzyme specificity of HDAC6 enzyme. The
detailed interaction showing important amino acids in ligand–receptor interactions
for RTS-V5 is shown in Fig. 2.4a. The hydroxamate moiety of RTS-V5 interacts with
the active site Zn2+ in monodentate fashion. Beyond the Zn2+ coordination, there are
other amino acids found important for its selectivity and specificity for HDAC6. The
aromatic ring of phenyl hydroxamate of the inhibitor is very close to the amino acid
F643. In the complex, it has been found that S531 forms hydrogen bond with the
inhibitor which may be important also for its selectivity. Lastly, typical bidentate
and monodentate interactions with different inhibitors are highlighted in Fig. 2.4b,
c, respectively.
From the analysis of different ligand–receptor complexes, it has been found that
the structure of the linker is important to maintain the orientation of the cap group
of the inhibitor toward the loop. The specific inhibitor of HDAC6 generally shows
important interaction with the amino acids like F583 and F643. In case of RTS-V5, it
has been found that the linker is very close to the amino acid F643. This interaction
is unique for HDAC6-specific inhibitor. In general, the binding site of HDAC6 is
wider than other class I HDACs. This feature allows the binding of inhibitors with
bulky cap groups as well as aromatic or heteroaromatic linker features to the active
site of HDAC6. This feature can be taken into account for the design of selective
HDAC6 inhibitor.
32 S. Khatun et al.

Fig. 2.4 a Ligand–receptor interaction for the inhibitor RTS-V5 with drHDAC6 CD2 binding
site (PDB: 6CW8). b Bidentate (PDB: 6DVO), c monodentate (PDB: 6PZO) interactions of the
inhibitors with the HDAC6 receptor

2.2.2 Insight into HDAC10 Crystal Structures

The HDAC10 crystal structure was determined at 2.85 Å resolutions for Y307F
zHDAC10 complexed with the trifluoromethyl ketone inhibitor. The structure has a
butterfly-like architecture where each domain adopted the α/β fold observed in other
HDAC proteins (Fig. 2.5). The structure is comprised amino terminal polyamine
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 33

Fig. 2.5 Structure of Y307F zHDAC10 complexed with the trifluoromethyl ketone inhibitor

deacetylase (PDAC) domain as well as the C-terminal pseudodeacetylase (𝚿DAC)

domain. PDAC domain of HDAC10 is catalytically active, whereas in case of
HDAC6, both domains are catalytically active. It was revealed that the tertiary struc-
ture of the PDAC domain is almost similar to the catalytic domain 1 and catalytic
domain 2 of HDAC6 [69–71]. The ligand binding site was situated at the base of
the active site tunnel of HDAC10 where the inhibitor bound in an extended confor-
mation. The trifluoromethyl ketone moiety of the inhibitor is making asymmetric
interactions with the Zn2+ ion in HDAC10 ligand binding site. Two histidine residues
H136 and H137 are making important hydrogen bond interactions with the receptor.
A close-up view of the active site of the enzyme HDAC10 revealed that the active
site is much constricted than other HDACs like HDAC6. In the active site, an amino
acid E274 is acting as gatekeeper residue and its electrostatic interactions with the
ligand may be responsible for specificity of the enzyme activity. Another structural
features present in the PDAC domain is that presence of 310 helix having a consensus
sequence P23 (E,A)CE26 sterically constricts the binding site of HDAC10. This allows
the binding of long slender polyamines in the HDAC10 binding site [2]. Thus, unique
structural features present in HDAC10 can be exploited for the design of the selective
HDAC10 inhibitors.
In order to get more structural insights with the catalytic mechanism involving
HDAC10, X-ray crystallographic study with the intact substrates into the active site
of “humanized” D. rerio (zebrafish) HDAC10 having A24E and D94A substitutions
was performed [63]. The structure gives important insights into substrate recognition
process in HDAC10 as well as stabilization mechanism of transition states in the
catalysis process. The studies highlight the importance of Y307 to assist the Zn2+
ion in polarizing the substrate carbonyl and stabilize the negative charge in transition
state complexes [70]. Recently, the crystal structure of HDAC10-Tubastatin A was
solved at 2.00 Å resolutions (Fig. 2.6). The structure shows that the hydroxamate
moiety of Tubastatin A forms complex with the Zn2+ ion in the active site of HDAC10
[71]. There is also a specific hydrogen bond formation with the ligand carbonyl group
to the Y307. The important histidine’s H136 and H137 form important hydrogen
34 S. Khatun et al.

Fig. 2.6 Crystal structure of HDAC10-tubastatin A complex

bond also with the ligand. The role of the histidine dyad in the HDAC10 reaction
mechanism bears some resemblance to that of HDAC6.
The phenyl group of the ligand forms aromatic interaction with the W205 and
F146. However, this aromatic interaction is not contributing to the selectivity of
HDAC10 as similar aromatic cavity is also seen in case of HDAC6. The tricyclic
tetrahydro-γ-carboline group present in the ligand acts as a capping group of
HDAC10 inhibitors. This is mainly interacting with the indole moiety of W205. E24
and E274 form important electrostatic interaction with the ligand as shown in Fig. 2.6.
These interactions will be very helpful to guide the selective HDAC10 inhibitors.
However, there is no structure available for human HDAC10 with the inhibitors,
which may guide more effectively the design of selective HDAC10 inhibitors.

2.3 Different Tools of in Silico Drug Discovery and Its

Applications

Numerous CADD applications are utilized at almost early phases of the drug
discovery cascades. Thus, CADD can be described as a method to accelerate and
economize the method of the drug development process [61, 72–75]. It allows better
engaging on experiments and subsequently reduces the cost as well as time to find
new drugs. CADD comprises (i) in silico design and prediction of novel compounds
by making the drug discovery and development process faster, (ii) identifying and
optimizing new compounds by the aid of computational approach, and (iii) eliminate
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 35

Fig. 2.7 Different tools of in silico drug discovery describing about structure- or ligand-based
approaches of lead optimization

compounds with undesirable properties and selecting candidates with more chances
for success.
Pharmacophore-based techniques currently are an integral part of many CADD
workflows (Fig. 2.7) [76–78] and have been extensively employed for many assign-
ments such as de novo design, virtual screening, and lead optimization. Phar-
macophore model can be generated from both receptor-based and ligand-based
techniques (Fig. 2.7).
Similarly, molecular docking [77, 81–85] and molecular dynamic (MD) simula-
tions allow to understand the three-dimensional binding mode of a given molecule
in the binding site of a macromolecule (protein/DNA). The binding affinity can
also be quantitatively predicted by a docking score, and the stability of the protein–
ligand complex is judged by proper MD simulation studies [86, 87]. More interest-
ingly, pharmacophore-based virtual screening when combined with docking analyses
provides great chance of acceptability [87].

2.3.1 Design Strategies for HDAC6 Inhibitors

To examine the potential of achieving several dimensions of isoform selectivity in

the inhibition of HDACs, Kozikowski et al. in 2008 synthesized a series of struc-
turally distinct HDAC inhibitors and applied QSAR modeling strategies to explain
the potency of HDAC6 inhibitory activities as well as its selectivity over HDAC1,
HDAC2, HDAC8, and HDAC10. The inhibitors are having the 2,4' -diaminobiphenyl
group, which is suitably decorated with an amino acid residue at the o-amino group
36 S. Khatun et al.

[88]. The amino acid acts as a potential isoform differentiating, surface recognition
element, and it is linked to a hydroxamate or mercaptoacetamide moieties that chelate
to the catalytic zinc ion.
Different significant QSAR models were developed individually for HDAC6,
HDAC1, HDAC2, HDAC8, and HDAC10. These models highlighted the importance
of lipophilicity (clogP) and indicator variables I-NHCOCH2SH, I-Thiazole for the
HDAC inhibitory activities. The result nicely explained the higher HDAC6 inhibitory
activities for phenylthiazoles (compound 2, compound 3) and lower HDAC6
inhibitory activities for biphenyl mercaptoacetamides (compound 1) (Fig. 2.8). The
cap group of the inhibitor is not contributing in a significant way for the inhibitory
activities as explained in Sect. 2.2.1. The QSAR models were also developed for
the selectivity of HDAC6 over other HDACs. These models explained the effects
of different structural and physicochemical properties of the inhibitors for its selec-
tive HDAC6 inhibition. In summary, these modeling strategies nicely correlated the
experimental and predicted HDAC6 activities. In addition, cell-based experiments
were carried out to determine the possible isoform and tissue selectivity of these
novel inhibitors. Finally, this study drew attention to the fact that certain mercaptoac-
etamides do show useful levels of HDAC6 selectivity. Most importantly, the current
research has identified two hydroxamates bearing meta-substituted phenylthiazole
CAPs (compound 2, compound 3) that have IC50 values < 0.2 nM in in vitro HDAC6
inhibition studies. Moreover, several phenylthiazoles were found to exhibit submi-
cromolar to low nanomolar IC50 values in the pancreatic cancer cell proliferation
studies.
Tang et al. used a combinatorial QSAR approach to build models for 59 chemically
diverse HDAC inhibitors [89]. The studies identified a novel HDAC6 inhibitor that
signifies the power of QSAR-based virtual screening strategies in HDAC-targeted
drug discovery. The variable selection methods of k nearest neighbor (kNN) and
support vector machines (SVM) are used in QSAR model building independently
by the use of Molconn Z and MOE chemical descriptors [90]. Highly predictive
QSAR models were developed with leave-one-out cross-validated (LOO-CV) q2 and
external R2 values as high as 0.80 and 0.87, respectively, utilizing the kNN/Molconn
Z approach. Extensive external validations on both kNN and SVM models were
conducted using two external datasets as described in Fig. 2.9. The Y-randomization
test was run in addition to external validation to determine the model’s robustness.
A rigorously validated QSAR models were used for virtual screening (VS) of an in-
house database collection of over 9.5 million compounds compiled from the ZINC7.0
database, the ASINEX Synergy libraries, the World Drug Index (WDI) database, and
other commercial databases. The study yielded 45 novel putative HDAC inhibitors.
These computational hits contained several unique structural features that were
absent in the original dataset. Four computational hits with interesting chemical
features were evaluated, out of which one compound was identified as selective
HDAC6 inhibitor (compound 4).
Zhao et al. developed the models of two classes of HDAC inhibitors (HDAC1 and
HDAC6) in 2013 [90]. The selectivity and activity of HDAC inhibitors were studied
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 37

Fig. 2.8 Model structure containing a ZBG, linker, and a CAP region with end groups for surface
recognition and isoform selectivity for HDAC6 inhibition. Compound 1 contains a biphenyl cap
and it is linked to a mercaptoacetamide moiety that chelates to the catalytic zinc ion. Compounds
2 and 3 are two hydroxamates bearing meta-substituted phenylthiazole CAPs that exhibit < 0.2 nM
IC50 values in the in vitro HDAC6 inhibition

using a two-step modeling approach. A schematic representation of the novel QSAR

approach is depicted in Fig. 2.10.
First, a binary classification model was built to classify two types of inhibitors
based on their activity against HDAC1 and HDAC6. Then, for each subclass, two
continuous models were created to predict the activity value of HDAC1 and HDAC6
inhibitors. All three models were developed using the GA-kNN method and dragon
descriptors. External validation was performed using an external prediction set and
Y-randomization test. For each of the three datasets, highly predictive models were
constructed. The classification accuracies of the models for the external test set
were as high as 100% for the classification model. External R2 values for HDAC1
and HDAC6 inhibitor consecutive models were 0.947 and 0.911, respectively. The
outcomes validated the models’ accuracy. All of the models were used to screen
1,000 compounds from the PubMed dataset. Virtual screening yielded 13 structurally
diverse consensus hits as HDAC6 inhibitors.
Pham-The et al. in 2017 explored diverse machine learning (ML) techniques for
the development of reliable QSAR models capable of distinguishing HDAC6 to
38 S. Khatun et al.

Fig. 2.9 Identification of novel inhibitor (compound 4) for HDAC6 by QSAR modeling of known
inhibitors, virtual screening, and experimental validation

HDAC2 inhibitors [91]. The ChEMBL (https://www.ebi.ac.uk/chembl/) and Drug-

Bank databases were used to curate a large, structurally diverse collection of chem-
icals. The database contains 191 compounds as HDAC6 inhibitor/HDAC2 non-
inhibitor and 95 compounds as HDAC6 non-inhibitor/HDAC2 inhibitor. The study
pointed out several important compounds such as quinazoline-4-one derivatives
(5), tetrahydro-1H-benzazepines (6), biphenyl hydroxylpyridin-2-thiones (7), 3-
hydroxypyridine-2-thiones (8), and phenyl hydroxamic acids (9) as potential HDAC6
inhibitors (Fig. 2.11).
Zeb et al. in 2018 designed a study to investigate non-hydroxamate HDAC6
inhibitors [92]. Ligand-based pharmacophore was established from a training set
of 26 compounds of HDAC6 inhibitors. A lowest total cost of 115.63, highest cost
difference of 135.00, lowest RMSD of 0.70, and highest correlation of 0.98 were the
statistical parameters of pharmacophore (Hypo1). Fischer’s randomization and test
set validation methods were used to validate the pharmacophore, which was then used
as a screening tool for chemical databases. The pharmacophore model (Fig. 2.12)
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 39

Fig. 2.10 Workflow assigned for two-step QSAR approach

5 6 7

8 9

Fig. 2.11 Different HDAC6-selective inhibitors pointed out in the study. The colored region
indicates important scaffolds like quinazoline-4-one moiety (pink), tetrahydro-1H-benzazepine
(blue), biphenyl hydroxylpyridin-2-thiones (green), 3-hydroxypyridine-2-thione (red), and phenyl
hydroxamic acid (sky blue) present in compounds 5, 6, 7, 8, and 9, respectively

indicates four features like HBA, HBD, RA, and HYP which are important for the
design of HDAC6 inhibitor.
The pharmacophore-based screening methods were applied to identify novel
HDAC6 inhibitors. To identify drug-like compounds, the screened compounds
were analyzed using fit value (> 10.00), estimated Inhibitory Concentration (IC50 )
(< 0.459), Lipinski’s Rule of Five, and ADMET Descriptors. In addition, the drug-
like hit compounds were docked into the active site of HDAC6 (PDB ID: 5EDU)
using GOLD software. The best docked compounds were selected on the basis of
goldfitness score > 66.46 and chemscore < 28.31 and hydrogen bonds with catalytic
40 S. Khatun et al.

HBD

9.598 2.278
3.717
10.657

7.202 3.689
HYP

HBA

RA

Fig. 2.12 Manual representation of pharmacophore model. The model consists of one HBA (blue),
one HBD (green), one RA (magenta), and one HYP (purple) as important pharmacophoric features

active residues. The binding modes of the final three hit compounds were investi-
gated by also using a 20-ns MD simulation. The MD simulation results showed that
the hit compounds formed several interactions including π-π—stacking, hydrogen
bonds, π-cation, π-sulfur, and hydrophobic interactions with the active site residues
of HDAC6. Furthermore, docking analysis was used to assess the proposed speci-
ficity of the newly discovered hits against HDAC8. The final hit molecules (10, 11,
and 12) have been proposed as promising platforms for the development of novel
HDAC6 inhibitors (Fig. 2.13).
Debnath et al. generated a number of five featured pharmacophore hypotheses
to identify selective HDAC6 inhibitors [93]. The study involved a combination
of pharmacophore-based virtual screening, molecular docking, 3D-QSAR, absorp-
tion, distribution, metabolism, excretion, and toxicity (ADMET) study, and in vitro
HDAC6 inhibitory activity assay of identified hits. Thirty-two known HDAC
inhibitors were selected from the literature and a common pharmacophore hypoth-
esis was generated. The best hypotheses ADDRR4 was composed of five features:
one hydrogen bond acceptors (A2), two hydrogen bond donors (D3, D4), and two
aromatic rings (R7 and R8). The 3D-QSAR model developed from the pharma-
cophore ADDRR4 was utilized to search the Phase database. The ligand pharma-
cophore mapping process was used to identify compounds in the database that shared
at least four pharmacophoric features. This step yielded 500 top-scoring hits with
fitness scores ≥ 1.0. These hit molecules were subjected to ADME filtration followed
by molecular docking against HDAC6 (PDB ID: 5WPB) to predict the binding
affinity toward the HDAC6. Generated pharmacophores were employed to match the
final five hits. In vitro HDAC inhibitory activity clearly demonstrated that compound
13 showed marginal selectivity for HDAC6 (IC50 = 0.62 nM). It was investigated
that the hit compound 13 preferentially binds to the catalytic domain 2 rather than
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 41

890 HDAC6 inhibitors

74 compounds
training set test set
(26 compounds) (48 compounds)

Both were classified into 3 groups:

Highly active Moderately active inactive
(IC50 <=100 nM/L) (100 IC50 <=10; 000 nM/L (IC50 >=10; 000 nM/L)

3D-QSAR PHARMACOPHORE MODEL GENERATION

Fischer's Randomization and Test Set validation,

Virtual screening
•Screening of chemical DBs
• Estimation of drug-like properties & ADMET Descriptors

Molecular docking Drug like hit compounds

and dynamics

Identified 3 hits as potential HDAC6 inhibitors

10 11 12

Fig. 2.13 Investigation of non-hydroxamate-based HDAC6 inhibitors using pharmacophore

modeling, molecular docking, and molecular dynamics simulation approach
42 S. Khatun et al.

the zinc finger domain. Figure 2.14 depicts the workflow for the identification of
novel HDAC6 inhibitor compound 13.
Sharma et al. carried out a 3D-QSAR study on structurally diverse HDAC6
inhibitors [94]. The inhibitory activities of the diverse set of ligands were corre-
lated with different field descriptors (steric, electrostatic, hydrophobic, donor, and
acceptor) by using common feature-based pharmacophore alignment for the 3D-
QSAR CoMFA, CoMSIA, docking, and MD simulations. The HipHop module in
Discovery Studio (DS v2.5) was used to select and align a highly selective dataset
of HDAC6 inhibitors, and the best model with the highest rank score was chosen for
CoMFA and CoMSIA model building. Internal and external validation methods were
used to validate the models. The models had a good statistical significance and a high
correlation of 0.978 and 0.991 for the best CoMFA and CoMSIA models, respec-
tively. These models could be used to predict the activities of the test set compounds
as well as to derive useful information about the molecules’ steric, electrostatic,
and hydrophobic properties. Further, the training and test set molecules were then
docked into the HDAC6 active site to correlate the results of 3D-QSAR with the
molecular interactions within the binding site of HDAC6. In addition to this, molec-
ular dynamics was used to indicate structural requirements for new inhibitor design.
The docked poses were compared with the contour maps obtained from CoMFA/
CoMSIA. For HDAC6 inhibitory action, the contour map studies revealed the rela-
tive relevance of steric, electrostatic, donor, and acceptor fields. Figure 2.15 depicts
the pharmacophoric characteristics that are critical for the selectivity and potency of
HDAC6 inhibitory action. This work confirms the soundness and predictive power
of the proposed models with excellent consistency between molecular docking and

32 HDAC 3D QSAR Pharmacophore based virtual screening

inhibitors modeling

500 Hits (Fitness score>1.0)

ADME filtration

Molecular
docking with
HDAC6

E-Pharmacophore matching
of best 5 hits
13
HDAC6 IC50 = 0.62 nM
In vitro testing of 5 selected
Novel HDAC6 selective inhibior hits against HDAC6

Fig. 2.14 Screening workflow for the novel HDAC6-selective inhibitor 13. Inhibitory activity of
compound 13 against HDAC6 was 0.62 nM
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 43

Fig. 2.15 Pharmacophoric Electron donor group

features essential for HDAC6
inhibitory activity [94]

Participate in
non-polar interactions

Sterically bulky
scaffold

Hydrophobic ring is
required for activity
Electron donor
group

Co-ordinates with Zn2+

CoMFA and CoMSIA contour maps and the also provides valuable insights for
a deeper comprehension of the basic physicochemical and structural features for
designing novel HDAC6 inhibitors for drug development.
To define structural motifs which aid HDAC6 selectivity, Ruzic et al. developed
valid and reliable 3D-QSAR models in order to define structural features that influ-
ence the inhibitory potency of hydroxamic acid-based HDAC1 and HDAC6 inhibitors
[95]. Virtual docking studies using crystal structures of human HDAC6 isoforms were
performed for conformer generation of dataset compounds. An attempt was made to
eliminate bias in conformer synthesis by employing geometrical optimization, and
as a result, using this innovative approach, descriptors for 3D-QSAR study related
to virtually bioactive conformations were developed. The obtained conformers were
utilized for generation of specific molecular descriptors (Grid-Independent Descrip-
tors–GRIND) and 3D-QSAR modeling. A comparative analysis of the descriptors
was used, to select specific structural determinants of the compounds for selec-
tive HDAC6 inhibition. GRIND descriptors were utilized using a combination of
ligand-based and fragment-based approaches to define the differences in structural
determinants critical for HDAC6 inhibition, as well as to provide directions for
designing novel HDAC6-selective inhibitors. The potential new ligands were chosen
for further research based on their projected HDAC6 selectivity, pharmacokinetic
profile, synthetic tractability, and in silico cytotoxicity against a variety of human
cancer cell lines. TIP-TIP GRIND variables were used to define the overall molecular
shape of the HDAC inhibitors. The previously stated variables agreed with the DRY-
O GRIND variable (var303) in the HDAC6 3D-QSAR model, indicating important
hydrophobic interactions of the CAP region with the outside rim of HDAC6 second
catalytic pocket. These variables showed that the distance between the CAP group
and the hydroxamic acid (linker length) should be increased (number of carbon
atoms equal or higher than 6). Fragment-based approach indicated 14 compounds
44 S. Khatun et al.

H500
S568 H499 G619
P501 H651
H499 P748
F620
F620 Y782
P501 G619

H500

Y782 S568

14 15

HDAC6 %inhibition (20μM)= 81% HDAC6 %inhibition (20μM)= 61%

Fig. 2.16 Structures of highly selective HDAC6 inhibitors 14 and 15

with increased selectivity toward HDAC6 isoform. The leverage values were used to
populate the novel developed compounds into the expected applicability domains,
and hence, the pKi values of the designed compounds were regarded accurately
predicted. Finally, the proposed compounds were chosen for synthesis and in vitro
testing because they had a superior anticipated HDAC6 selectivity profile, enhanced
in silico ADME properties, and lower in silico toxicological concerns than clinically
approved HDACi. The rational drug design technique used in this study would aid
in the search for new chemically appealing CAP groups that could behave as unique
interaction groups with the outer rim of HDAC6 CD2 binding pocket.
In order to discover novel inhibitor for targeting HDAC6 selectively for clinical
applications, Yan et al. [96] employed virtual screening methods through the pharma-
cophore generation and pharmacophore-based virtual screening, molecular docking,
PAINS remover, and MD simulations. Following virtual screening, 15 molecules
were chosen for biological activity research. Two of the hits (compounds 14 and
15) demonstrated inhibitory efficacy against HDAC6 after in vitro bioassays, with
compound 14 inhibiting HDAC6 by 81% at a dose of 20 μM. Furthermore, inhibitory
efficacy showed that these compounds were highly selective to HDAC6 (Fig. 2.16).
The investigation of binding modalities of these hit molecules can provide a point of
reference for the future development of highly active HDAC6 inhibitors.

2.4 Design Strategies for HDAC 10 Inhibitors

Histone deacetylase 10 (HDAC10) has been linked to the pathophysiology of several

malignancies and neurological illnesses, making the finding of new inhibitors of
the isoform a crucial undertaking. However, the lack of a crystallographic structure
of human HDAC10 (hHDAC10) hinders efforts to develop structure-based drugs.
Furthermore, from the standpoints of medicinal chemistry and structure-based drug
design, HDAC10 has received little attention as a cancer therapeutic target so far.
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 45

Uba et al. in 2019 employed the X-ray crystal structure of Danio rerio (zebrafish)
HDAC10 (PDB ID: 5TD7) as a template structure to model the 3D structure of human
HDAC10 due to their high sequence similarity, especially around their catalytic
domain [97]. The best model (M0017) was evaluated by computing its z-score
(a measure of the deviation of the total energy of the structure with respect to
an energy distribution derived from random conformations) and by docking of a
series of known HDAC10 inhibitors to its catalytic cavity. Moreover, ligand-based
virtual screening against the ZINC database was used to identify potential HDAC10-
selective inhibitors. The screening identified the top 100 hits with scaffolds similar
to quisinostat. M0017, the top-ranked built homology model, and its complexes
with quisinostat, as well as the best hit, compound 16, were subjected to molecular
dynamics simulations. CHARMM-GUI server employing CHARMM36 force field
(http://www.charmm-gui.org/) was used to generate the input files for nanoscale
molecular dynamics (NAMD). M0017’s catalytic domain was thus extracted and
aligned to the catalytic domain of Danio rerio HDAC10’s experimental crystal struc-
ture. Comparative analysis of root mean-squared deviation (RMSD), root-mean-
squared fluctuation (RMSF), potential energy, and radius of gyration (Rg) of these
systems revealed that HDAC10 compound 16 complex remained the most stable
over time. Therefore, M0017 could be used as a structure-based inhibitor against
HDAC10, and compound 16 (Fig. 2.17) could be used as a lead compound and
a scaffold for further optimization in the direction of selective HDAC10 inhibitor
design.
In 2020, Uba et al. performed the first molecular dynamics study of hHDAC10
developed using the crystallographic structure of Danio rerio (zebrafish) HDAC10
(zHDAC10) as a template [98]. The objective behind exploring the two homologs,
despite the fact that they came from different organisms, was to allow for
robust comparisons that led to the inference of reliable hHDAC10 conforma-
tion from zHDAC10. Both hHDAC10 and zHDAC10, as well as their complexes

Y305
H135

H174
G143
Interactions
F144
Van der Waals
Hydrogen bonding
Pi-Cation/Pi-Anion
Pi-Pi T-shaped
Alkyl

A92
E272

W203

Fig. 2.17 Different interactions shown by hit compound 16 in the active site of HDAC10
46 S. Khatun et al.

with trichostatin A (TSA), quisinostat, and the natural ligand (5TD7), 7-[(3-
aminopropyl)amino]-1,1,1-trifluoroheptane-2,2-diol (PDB ID; FKS), were subjected
to 100 ns unrestrained MD simulations. Comparative analyses of the MD trajecto-
ries revealed that zHDAC10 and its complexes were more stable over time than
hHDAC10 and its associated complexes. Nonetheless, docking of active and inac-
tive set molecules demonstrated that more stable conformations of hHDAC10 might
be obtained over time [98]. The current findings suggest that the initial predicted
structure of hHDAC10 may not accurately reflect the active conformation in the
real biological environment. The experimentally determined crystallographic struc-
ture, zHDAC10, demonstrated greater structural stability than the homology-modeled
structure, hHDAC10. Further docking of inactive and active set molecules indicated
that the modeled structure could be utilized to develop structural based inhibitors with
confidence. Overall, this research could offer more information on the reliability of
the HDAC10-predicted structure for application in selective inhibitor design.
Geraldy et al. discovered tubastatin A, an HDAC6 inhibitor, to potently bind with
HDAC10 pocket [99]. Docking into human HDAC10 homology models revealed
that a hydrogen bond between cap group nitrogen atoms is responsible for strong
HDAC10 binding. The gatekeeper residue Glu272 is the single element that confers
HDAC10 PDAC over HDAC activity (Fig. 2.18). This selectivity is due to the polar
interaction between glutamic acid and a positively charged polyamine. Different
homology models of hHDAC10 were built and docking studies were performed to
investigate the binding of inhibitors into HDAC10 pocket. Initially, two hHDAC10
homology models Model I and Model II were prepared, based on a crystal structure
of zHDAC10 (PDB ID 5td7) and hHDAC6 (PDB ID 5edu), respectively. Docking
of tubastatin A was performed using both Model I and Model II. Model I failed to
demonstrate reliable docking poses, whereas Model II demonstrated high-scoring
docking poses of tubastatin A, with a hydrogen bond to Glu272, and has an L1
loop that is more similar to its parent hHDAC6 structure (Fig. 2.18). Further, to
investigate the loop flexibility, two additional models (Model III and Model IV)
were built. Surprisingly, docking of tubastatin A into these models was successful,
leading to hydrogen-bonded docking poses similar to what was found in Model II. To
summarize, the four models only revealed reliable docking poses when the L1 loop
adopted a conformation different from that of the zHDAC10 structure, indicating
that the loop flexibility of L1 is required for tubastatin A binding.
To accelerate the drug design efforts, Herbst-Gervasoni et al. described the
preparation of “humanised” zebrafish HDAC10, which has two amino acid substi-
tutions, A24E and D94A that provide an active site outline more comparable to
that of human HDAC10 [71]. The X-ray crystal structures of this HDAC10 variant
complexed with tubastatin A and indole analogs with pendant tertiary amines show
that inhibitors capable of hydrogen bonding with gatekeeper E274 have a high affinity
and selectivity for HDAC10 over HDAC6 [71] (Fig. 2.19).
Ukey et al. used computational methods such as homology modeling, docking,
pharmacophore analysis, and molecular dynamic (MD) simulations to decipher
the specificity rendering interaction features of HDAC 10 and other ZnHDACs
[100]. Eight approved pan-HDAC inhibitors (PHIs) and experimentally validated 91
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 47

G143 D172

D265

A92

Y305

E272

Fig. 2.18 Docking pose of tubastatin A in hHDAC10

Y307
Zn
E274

H137

P23
F146
E24

Fig. 2.19 Interaction of HDAC10 (Pdb: 6wdv) with indole analogue

reported subtype-specific inhibitors (SSIs) (activity/1000 nM) of various ZnHDACs

were docked to 40 MD generated conformations of each ZnHDAC after which MM-
GBSA binding energy estimates were made. To pinpoint distinct subtype-specific
interaction features for each Class II ZnHDAC, the sequences, structures, phys-
iochemical properties, and patterns of interaction of the binding sites procured
from docking were thoroughly compared. A 20-ns MD simulation was run on 12
complexes (each Class II ZnHDAC bound to one SSI and one PHI) in explicit water
models to further confirm the stabilities of these features. Despite high sequence simi-
larity, different pharmacophoric patterns were seen in the binding pockets of each
subtype. Due to subtype-specific protein–ligand interactions, the presence of amides,
ketone, hydroxyl, carboxyl groups, moieties occupying additional sub-pockets and
interacting with Zn2+ , etc., in the SSIs affects the orientations of the binding site
48 S. Khatun et al.

H135
P132
Y305 H134
E302 F144
L307

P271 V173
R27
E26 H174
I25 F202
E272 H175
W203

Fig. 2.20 Interactions (green: π-stacking; red: cation-π; magenta: H-bond) of SSIs and PHIs with
the relevant class II ZnHDACs BS residues

residues (BSRs). Such distinct interaction features and pharmacophoric patterns

(Fig. 2.20) can be used to develop subtype-specific ZnHDAC inhibitors.

2.5 Conclusion

HDAC6 is the only HDAC known to have two catalytic domains and to be predom-
inately cytoplasmic in its structure. This distinguishing feature allowed HDAC6 to
target specific substrates involved in proteasomal degradation, cell shape and migra-
tion, microtubule dynamics, apoptosis, axonal growth abnormalities, and different
signaling pathways that contribute to the pathogenic response of many diseases. This
broad variety of roles and activity of HDAC6 is well employed in many malignancies,
neurological illnesses, epigenetic rare diseases, and inflammatory diseases. Utilizing
a variety of selective HDAC6i, various researches have been published to date exam-
ining and comprehending the cellular and physiological roles of HDAC6. Although
numerous studies described the preclinical testing of HDAC6 isoform inhibitors, only
ACY-1215 (ricolinostat) and ACY-241 (citarinostat) (Fig. 2.21) have so far entered
phase II clinical trials. Designing efficient and selective HDAC6i must undoubt-
edly take druggability and poor bioavailability into consideration. Since the bulky
cap group is made up of extremely hydrophobic moieties and the linker function is
mostly made up of linear elongation alkyl functions, the hydrophobicity combined
with the flexibility provided by linear functionalities may result in varying interaction
with HDAC enzymes.
Like HDAC6, HDAC10 demonstrates a variety of intricate functions and clin-
ical implications. The theoretical groundwork for clinical application is laid by its
functions, which include DNA damage repair, autophagy, gene transcription and
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 49

ACY-241 (Citarinostat)

ACY-1215 (Ricolinostat)

Fig. 2.21 Structures of HDAC6-selective inhibitors that have entered phase II clinical trials

HDAC10-mediated tumor cell proliferation, migration, invasion, apoptosis, metas-

tasis, angiogenesis, and immunological modulation. There are still difficulties to
be overcome, though. In order to provide a more comprehensive understanding of
HDAC10 and other class IIb enzyme, further information on the mechanisms of
aberrant HDAC regulation is first and foremost necessary. HDAC10 may become
a crucial anti-tumor target and contribute to therapeutic applications with more
in-depth research.
So far, the majority of HDAC6- and HDAC10-selective inhibitors have hydrox-
amic acids as ZBG with different structural changes on the cap and linker regions.
Despite the fact that numerous other ZBGs with higher potencies have been identified,
none of them displayed substantial cellular activity. Nonetheless, the greater zinc-
binding ability of various ZBGs, such as hydroxamate, may be a serious issue in terms
of selective binding to specific HDAC, resulting in undesired toxicities. As a result,
the main difficulty hampered by these functions is the selectivity of HDAC Class IIb
inhibitors. Moreover, these moieties are not only responsible for enzyme binding,
but also for druggability and bioavailability. Aside from that, crystallographic data
can be used to examine the binding mode of contacts, which can assist to discover
essential functional aspects as well as vital amino acid residues. These are useful
in the development of selective HDACi. Furthermore, the time and effort required
in designing selective inhibitors prior to synthesis and biological screening may be
lessened by the initial ligand-docking interactions with the inhibitors of HDAC6 and
10 followed by ADMET screening using various modeling techniques. Taking these
critical features into account, it is anticipated that selective and effective HDAC6 and
HDAC10 specific inhibitors may be designed in future.
Overall, this chapter outlines the distinctive structural features of class IIb HDACs,
as well as their varied interactions with inhibitors. This would assist in designing and
50 S. Khatun et al.

developing more effective and selective class IIb HDACi, overcoming current hurdles
and broadening their clinical perspectives.

Conflict of Interest Authors do not have any conflict of interest.

Acknowledgements Authors sincerely acknowledge the Department of Pharmaceutical Tech-

nology, Jadavpur University, Kolkata, India, for providing the research facilities.

References

1. Wang Y, Xie Q, Tan H, Liao M, Zhu S, Zheng LL, Huang H, Liu B (2021) Targeting cancer
epigenetic pathways with small-molecule compounds: therapeutic efficacy and combination
therapies. Pharmacol Res. https://doi.org/10.1016/j.phrs.2021.105702
2. Prachayasittikul V, Prathipati P, Pratiwi R, Phanus-Umporn C, Malik AA, Schaduangrat N,
Seenprachawong K, Wongchitrat P, Supokawej A, Prachayasittikul V, Wikberg JE, Nantase-
namat C (2017) Exploring the epigenetic drug discovery landscape. Expert Opin Drug Discov
12:345–362. https://doi.org/10.1080/17460441.2017.1295954
3. Miranda Furtado CL, Dos Santos Luciano MC, Silva Santos RD, Furtado GP, Moraes
MO, Pessoa C (2019) Epidrugs: targeting epigenetic marks in cancer treatment. Epigenetics
14:1164–1176. https://doi.org/10.1080/15592294.2019.1640546
4. https://www.nature.com/scitable/topicpage/epigenetic-influences-and-disease-895/.
Accessed on 19 June 2021
5. Yu B, Ouyang L (2019) Epigenetic regulation and drug discovery for cancer therapy. Curr
Top Med Chem 19:971. https://doi.org/10.2174/156802661912190730153906
6. Shetty MG, Pai P, Deaver RE, Satyamoorthy K, Babitha KS (2021) Histone deacetylase 2
selective inhibitors: a versatile therapeutic strategy as next generation drug target in cancer
therapy. Pharmacol Res 170:105695. https://doi.org/10.1016/j.phrs.2021.105695
7. Nebbioso A, Tambaro FP, Dell’Aversana C, Altucci L (2018) Cancer epigenetics: moving
forward. PLoS Genet 14:e1007362. https://doi.org/10.1371/journal.pgen.1007362
8. Rasool M, Malik A, Naseer MI, Manan A, Ansari SA, Begum I, Qazi MH, Pushparaj PN,
Abuzenadah AM, Al-Qahtani MH, Kamal MA, Gan SH (2015) The role of epigenetics in
personalized medicine: challenges and opportunities. BMC Med Genomics 8:S5. https://doi.
org/10.1186/1755-8794-8-S1-S5
9. Biel M, Wascholowski GA (2005) Epigenetics—an epicenter of gene regulation: histones
and histone-modifying enzymes. Angew Chem Int Ed 44:3186–3216. https://doi.org/10.1002/
anie.200461346
10. Glozak MA, Sengupta N, Zhang X, Seto E (2005) Acetylation and deacetylation of non-
histone proteins. Gene 363:15–23. https://doi.org/10.1016/j.gene.2005.09.010
11. Lawlor L, Yang XB (2019) Harnessing the HDAC-histone deacetylase enzymes, inhibitors
and how these can be utilised in tissue engineering. Int J Oral Sci 11:20. https://doi.org/10.
1038/s41368-019-0053-2
12. Hull EE, Montgomery MR, Leyva KJ (2016) HDAC inhibitors as epigenetic regulators of
the immune system: impacts on cancer therapy and inflammatory diseases. Biomed Res Int
2016:8797206. https://doi.org/10.1155/2016/8797206
13. Amin SA, Trivedi P, Adhikari N, Routholla G, Vijayasarathi D, Das S, Ghosh B, Jha T (2021)
Quantitative activity–activity relationship (QAAR) driven design to develop hydroxamate
derivatives of pentanoic acids as selective HDAC8 inhibitors: synthesis, biological evaluation
and binding mode of interaction studies. New J Chem 45:17149. https://doi.org/10.1039/d1n
j02636d
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 51

14. Bourguet E, Ozdarska K, Moroy G, Jeanblanc J, Naassila M (2018) Class I HDAC inhibitors:
potential new epigenetic therapeutics for alcohol use disorder (AUD). J Med Chem 61:1745–
1766. https://doi.org/10.1021/acs.jmedchem.7b00115
15. Wang N, Peng YJ, Su X, Prabhakar NR, Nanduri J (2021) Histone deacetylase 5 is an early
epigenetic regulator of intermittent hypoxia induced sympathetic nerve activation and blood
pressure. Front Physiol 12:688322. https://doi.org/10.3389/fphys.2021.688322
16. Cao H, Li L, Yang D, Zeng L, Yewei X, Yu B, Liao G, Chen J (2019) Recent progress in histone
methyltransferase (G9a) inhibitors as anticancer agents. Eur J Med Chem 179:537–546. https:/
/doi.org/10.1016/j.ejmech.2019.06.072
17. Vaidya GN, Rana P, Venkatesh A, Chatterjee DR, Contractor D, Satpute DP, Nagpure M,
Jain A, Kumar D (2021) Paradigm shift of “classical” HDAC inhibitors to “hybrid” HDAC
inhibitors in therapeutic interventions. Eur J Med Chem 209:112844. https://doi.org/10.1016/
j.ejmech.2020.112844
18. Amin SA, Adhikari N, Jha T (2017) Is dual inhibition of metalloenzymes HDAC-8 and MMP-
2 a potential pharmacological target to combat hematological malignancies? Pharmacol Res
122:8–19. https://doi.org/10.1016/j.phrs.2017.05.002
19. Amin SA, Adhikari N, Jha T (2017) Structure-activity relationships of hydroxamate-based
histone deacetylase-8 inhibitors: reality behind anticancer drug discovery. Future Med Chem
9:2211–2237. https://doi.org/10.4155/fmc-2017-0130
20. Atadja P (2009) Development of the pan-DAC inhibitor panobinostat (LBH589): successes
and challenges. Cancer Lett 280:233–241. https://doi.org/10.1016/j.canlet.2009.02.019
21. Osko JD, Christianson DW (2020) Structural determinants of affinity and selectivity in the
binding of inhibitors to histone deacetylase 6. Bioorg Med Chem Lett 30:127023. https://doi.
org/10.1016/j.bmcl.2020.127023
22. Rajan PK, Udoh UA, Sanabria JD, Banerjee M, Smith G, Schade MS, Sanabria J, Sodhi K,
Pierre S, Xie Z, Shapiro JI, Sanabria J (2020) The role of histone acetylation-/methylation-
mediated apoptotic gene regulation in hepatocellular carcinoma. Int J Mol Sci 21:8894. https:/
/doi.org/10.3390/ijms21238894
23. de Ruijter AJM, van Gennip AH, Caron HN, Kemp S, van Kuilenburg ABP (2003) Histone
deacetylases (HDACs): characterization of the classical HDAC family. Biochem J 370:737–
749. https://doi.org/10.1042/bj20021321
24. Wang X-X, Wan R-Z, Liu Z-P (2018) Recent advances in the discovery of potent and selective
HDAC6 inhibitors. Eur J Med Chem 143:1406–1418. https://doi.org/10.1016/j.ejmech.2017.
10.040
25. Marmorstein R, Roth SY (2001) Histone acetyltransferases: function, structure, and catalysis.
Curr Opin Genet Dev 11:155–161. https://doi.org/10.1016/s0959-437x(00)00173-8
26. Haberland M, Montgomery RL, Olson EN (2009) The many roles of histone deacetylases in
development and physiology: implications for disease and therapy. Nat Rev Genet 10:32–42.
https://doi.org/10.1038/nrg2485
27. Taunton J, Hassig CA, Schreiber SL (1996) A mammalian histone deacetylase related to the
yeast transcriptional regulator Rpd3p. Science 272:408–411. https://doi.org/10.1126/science.
272.5260.408
28. Yang WM, Inouye C, Zeng Y, Bearss D, Seto E (1996) Transcriptional repression by YY1
is mediated by interaction with a mammalian homolog of the yeast global regulator RPD3.
PNAS 93:12845–12850. https://doi.org/10.1073/pnas.93.23.12845
29. Yang WM, Yao YL, Sun JM, Davie JR, Seto E (1997) Isolation and characterization of cDNAs
corresponding to an additional member of the human histone deacetylase gene family. J Biol
Chem 272:28001–28007. https://doi.org/10.1074/jbc.272.44.28001
30. Hu E, Chen Z, Fredrickson T, Zhu Y, Kirkpatrick R, Zhang GF, Johanson K, Sung CM, Liu R,
Winkler J (2000) Cloning and characterization of a novel human class I histone deacetylase
that functions as a transcription repressor. J Biol Chem 275:15254–15264. https://doi.org/10.
1074/jbc.M908988199
31. Kao HY, Downes M, Ordentlich P, Evans RM (2000) Isolation of a novel histone deacetylase
reveals that class I and class II deacetylases promote SMRT-mediated repression. Genes Dev
14:55–66
52 S. Khatun et al.

32. Kao HY, Lee CH, Komarov A, Han CC, Evans RM (2002) Isolation and characterization of
mammalian HDAC10, a novel histone deacetylase. J Biol Chem 277:187–193. https://doi.
org/10.1074/jbc.M108931200
33. Zhou X, Richon VM, Rifkind RA, Marks PA (2000) Identification of a transcriptional
repressor related to the noncatalytic domain of histone deacetylases 4 and 5. Proc Natl Acad
Sci 97:1056–1061. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC15519/. Accessed 2 May
2020
34. Seto E, Yoshida M (2014) Erasers of histone acetylation: the histone deacetylase enzymes.
Cold Spring Harb Perspect Biol 6:a018713. https://doi.org/10.1101/cshperspect.a018713
35. Gao L, Cueto AF, Atadja P (2002) Cloning and functional characterization of HDAC11,
a novel member of the human histone deacetylase family. J Biol Chem 277:25748–25755.
https://doi.org/10.1074/jbc.M111871200
36. Finnin MS, Donigian JR, Cohen A, Richon VM, Rifkind RA, Marks PA, Breslow R, Pavletich
NP (1999) Structures of a histone deacetylase homologue bound to the TSA and SAHA
inhibitors. Nature 401:188–193. https://doi.org/10.1038/43710
37. Wang DF, Helquist P, Wiech NL, Wiest O (2005) Toward selective histone deacetylase
inhibitor design: homology modeling, docking studies, and molecular dynamics simulations
of human class I histone deacetylases. J Med Chem 48:6936–6947. https://doi.org/10.1021/
jm0505011
38. Adhikari N, Amin SA, Trivedi P, Jha T, Ghosh B (2018) HDAC3 is a potential validated
target for cancer: an overview on the benzamide-based selective HDAC3 inhibitors through
comparative SAR/QSAR/QAAR approaches. Eur J Med Chem 157:1127–1142. https://doi.
org/10.1016/j.ejmech.2018.08.081
39. Grozinger CM, Hassig CA, Schreiber SL (1999) Three proteins define a class of human
histone deacetylases related to yeast Hda1p. Proc Natl Acad Sci 96:4868–4873. https://doi.
org/10.1073/pnas.96.9.4868
40. Li Y, Shin D, Kwon SH (2013) Histone deacetylase 6 plays a role as a distinct regulator of
diverse cellular processes. FEBS J 280:775–793. https://doi.org/10.1111/febs.12079
41. Hubbert C, Guardiola A, Shao R, Kawaguchi Y, Ito A, Nixon A, Yoshida M, Wang XF, Yao
TP (2002) HDAC6 is a microtubule-associated deacetylase. Nature 417:455–458. https://doi.
org/10.1038/417455a
42. Shen S, Svoboda M, Zhang G, Cavasin MA, Motlova L, McKinsey TA, Eubanks JH, Bařinka
C, Kozikowski AP (2020) Structural and in vivo characterization of tubastatin A, a widely
used histone deacetylase 6 inhibitor. ACS Med Chem Lett 11:706–712. https://doi.org/10.
1021/acsmedchemlett.9b00560
43. Simões-Pires C, Zwick V, Nurisso A, Schenker E, Carrupt PA, Cuendet M (2013) HDAC6 as
a target for neurodegenerative diseases: what makes it different from the other HDACs? Mol
Neurodegener 8:1–16. https://doi.org/10.1186/1750-1326-8-7
44. Li T, Zhang C, Hassan S, Liu X, Song F, Chen K, Zhang W, Yang J (2018) Histone deacetylase
6 in cancer. J Hematol Oncol 11:1–10. https://doi.org/10.1186/s13045-018-0654-9
45. Brindisi M, Saraswati AP, Brogi S, Gemma S, Butini S, Campiani G (2020) Old but gold:
tracking the new guise of histone deacetylase 6 (HDAC6) enzyme as a biomarker and thera-
peutic target in rare diseases. J Med Chem 63:23–39. https://doi.org/10.1021/acs.jmedchem.
9b00924
46. Osko JD, Porter NJ, Narayana Reddy PA, Xiao YC, Rokka J, Jung M, Hooker JM, Salvino JM,
Christianson DW (2020) Exploring structural determinants of inhibitor affinity and selectivity
in complexes with histone deacetylase 6. J Med Chem 63:295–308. https://doi.org/10.1021/
acs.jmedchem.9b01540
47. Cheng F, Zheng B, Wang J, Zhao G, Yao Z, Niu Z, He W (2021) Histone deacetylase 10,
a potential epigenetic target for therapy. Biosci Rep. 41:BSR20210462. https://doi.org/10.
1042/BSR20210462
48. Pojani E, Barlocco D (2022) Selective inhibitors of histone deacetylase 10 (HDAC-10). Curr
Med Chem 29:2306–2321. https://doi.org/10.2174/0929867328666210901144658
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 53

49. Morgen M, Steimbach RR, Géraldy M, Hellweg L, Sehr P, Ridinger J, Witt O, Oehme I,
Herbst-Gervasoni CJ, Osko JD, Porter NJ, Christianson DW, Gunkel N, Miller AK (2020)
Design and synthesis of dihydroxamic acids as HDAC6/8/10 inhibitors. ChemMedChem
15:1163–1174. https://doi.org/10.1002/cmdc.202000149
50. Oehme I, Lodrini M, Brady NR, Witt O (2013) Histone deacetylase 10-promoted autophagy
as a druggable point of interference to improve the treatment response of advanced
neuroblastomas. Autophagy 9:2163–2165. https://doi.org/10.4161/auto.26450
51. Lakshmaiah KC, Jacob LA, Aparna S, Lokanatha D, Saldanha SC (2014) Epigenetic therapy
of cancer with histone deacetylase inhibitors. J Cancer Res Ther 10:469–478. https://doi.org/
10.4103/0973-1482.137937
52. Bugide S, Gupta R, Green MR, Wajapeyee N (2021) EZH2 inhibits NK cell-mediated anti-
tumor immunity by suppressing CXCL10 expression in an HDAC10-dependent manner. Proc
Natl Acad Sci USA 118:e2102718118. https://doi.org/10.1073/pnas.2102718118
53. Muratov EN, Bajorath J, Sheridan RP, Tetko IV, Filimonov D, Poroiko V, Oprea TI, Baskin II,
Varnek A, Roitberg A, Isayev O (2020) QSAR without borders. Chem Soc Rev 49:3525–3564.
https://doi.org/10.1039/d0cs00098a
54. Amin SA, Gayen S (2016) Modelling the cytotoxic activity of pyrazolo-triazole hybrids
using descriptors calculated from the open source tool “PaDEL-descriptor.” J Taibah Univ
Sci 10:896–905. https://doi.org/10.1016/j.jtusci.2016.04.009
55. Lombardino JG, Lowe JA (2004) The role of the medicinal chemist in drug discovery–then
and now. Nat Rev Drug Discov 3:853–862. https://doi.org/10.1038/nrd1523
56. Hughes JP, Rees S, Kalindjian SB, Hughes PKL (2011) Principles of early drug discovery.
Br J Pharmacol 162:1239–1249. https://doi.org/10.1111/j.1476-5381.2010.01127.x
57. Bajorath J (2002) Integration of virtual and high-throughput screening. Nat Rev Drug Discov
1:882–894. https://doi.org/10.1038/nrd941
58. Amin SA, Adhikari N, Jha T, Gayen S (2016) First molecular modeling report on novel
arylpyrimidine kynurenine monooxygenase inhibitors through multi-QSAR analysis against
Huntington’s disease: a proposal to chemists! Bioorg Med Chem Lett 26:5712–5718. https:/
/doi.org/10.1016/j.bmcl.2016.10.058
59. Kiriiri GK, Njogu PM, Mwangi AN (2020) Exploring different approaches to improve the
success of drug discovery and development projects: a review. Futur J Pharm Sci 6:27. https:/
/doi.org/10.1186/s43094-020-00047-9
60. Shaker B, Ahmad S, Lee J, Jung C, Na D (2021) In silico methods and tools for drug discovery.
Comput Biol Med 137:104851. https://doi.org/10.1016/j.compbiomed.2021.104851
61. Radaeva M, Dong X, Cherkasov A (2020) The use of methods of computer-aided drug
discovery in the development of topoisomerase II inhibitors: applications and future directions.
J Chem Inf Model 60:3703–3721. https://doi.org/10.1021/acs.jcim.0c00325
62. Boyault C, Sadoul K, Pabion M, Khochbin S (2007) HDAC6, at the crossroads between
cytoskeleton and cell signaling by acetylation and ubiquitination. Oncogene 26:5468–5476.
https://doi.org/10.1038/sj.onc.1210614
63. Haggarty SJ, Koeller KM, Wong JC, Grozinger CM, Schreiber SL (2003) Domain-selective
small-molecule inhibitor of histone deacetylase 6 (HDAC6)-mediated tubulin deacetylation.
PNAS 100:4389–4394. https://doi.org/10.1073/pnas.0430973100
64. Zhang X, Yuan Z, Zhang Y, Yong S, Salas-Burgos A, Koomen J, Olashaw N, Parsons JT,
Yang XJ, Dent SR, Yao TP, Lane WS, Seto E (2007) HDAC6 modulates cell motility by
altering the acetylation level of cortactin. Mol Cell 27:197–213. https://doi.org/10.1016/j.mol
cel.2007.05.033
65. Han Y, Jeong HM, Jin YH, Kim YJ, Jeong HG, Yeo CY, Lee KY (2009) Acetylation of histone
deacetylase 6 by p300 attenuates its deacetylase activity. Biochem Biophys Res Commun
383:88–92. https://doi.org/10.1016/j.bbrc.2009.03.147
66. Hook SS, Orian A, Cowley SM, Eisenman RN (2002) Histone deacetylase 6 binds polyu-
biquitin through its zinc finger (PAZ domain) and copurifies with deubiquitinating enzymes.
PNAS 99:13425–13430. https://doi.org/10.1073/pnas.172511699.
54 S. Khatun et al.

67. Liu Y, Li L, Min J (2016) HDAC6 finally crystal clear. Nat Chem Biol 12:660–661. https://
doi.org/10.1038/nchembio.2158
68. Miyake Y, Keusch JJ, Wang L, Saito M, Hess D, Wang X, Melancon BJ, Helquist P, Gut
H, Matthias P (2016) Structural insights into HDAC6 tubulin deacetylation and its selective
inhibition. Nat Chem Biol 12:748–754. https://doi.org/10.1038/nchembio.2140
69. Hai Y, Shinsky SA, Porter NJ, Christianson DW (2017) Histone deacetylase 10 structure and
molecular function as a polyamine deacetylase. Nat commun 8:1–9. https://doi.org/10.1038/
ncomms15368
70. Herbst-Gervasoni CJ, Christianson DW (2021) X-ray crystallographic snapshots of substrate
binding in the active site of histone deacetylase 10. Biochem 60:303–313. https://doi.org/10.
1021/acs.biochem.0c00936
71. Herbst-Gervasoni CJ, Steimbach RR, Morgen M, Miller AK, Christianson DW (2020) Struc-
tural basis for the selective inhibition of HDAC10, the cytosolic polyamine deacetylase. ACS
Chem Biol 15:2154–2163. https://doi.org/10.1021/acschembio.0c00362
72. Amin SA, Jha T (2020) Fight against novel coronavirus: a perspective of medicinal chemists.
Eur J Med Chem 201:112559. https://doi.org/10.1016/j.ejmech.2020.112559
73. Amin SA, Adhikari N, Gayen S, Jha T (2019) Reliable structural information for rational
design of benzoxazole type potential cholesteryl ester transfer protein (CETP) inhibitors
through multiple validated modeling techniques. J Biomol Struct Dyn 37:4528–4541. https:/
/doi.org/10.1080/07391102.2018.1552895
74. Seidel T, Schuetz DA, Garon A, Langer T (2019) The pharmacophore concept and its appli-
cations in computer-aided drug design. Prog Chem Org Nat Prod 110:99–141. https://doi.org/
10.1007/978-3-030-14632-0_4
75. Macalino SJ, Gosu V, Hong S, Choi S (2015) Role of computer-aided drug design in modern
drug discovery. Arch Pharm Res 38:1686–1701. https://doi.org/10.1007/s12272-015-0640-5
76. Choudhury C, Sastry GN (2019) Pharmacophore modelling and screening: concepts,
recent developments and applications in rational drug design. In: Structural bioinformatics:
applications in preclinical drug discovery process, Springer, Cham, pp 25–53
77. Kontoyianni M (2017) Docking and virtual screening in drug discovery. Methods Mol Biol
1647:255–266. https://doi.org/10.1007/978-1-4939-7201-2_18
78. Voet A, Zhang KY (2012) Pharmacophore modelling as a virtual screening tool for
the discovery of small molecule protein-protein interaction inhibitors. Curr Pharm Des
18:4586–4598. https://doi.org/10.2174/138161212802651616
79. Yang SY (2010) Pharmacophore modeling and applications in drug discovery: challenges
and recent advances. Drug Discov Today 15:444–450. https://doi.org/10.1016/j.drudis.2010.
03.013
80. Kutlushina A, Khakimova A, Madzhidov T, Polishchuk P (2018) Ligand-based pharma-
cophore modeling using novel 3d pharmacophore signatures. Molecules 23:3094. https://
doi.org/10.3390/molecules23123094
81. Guedes IA, de Magalhães CS, Dardenne LE (2014) Receptor-ligand molecular docking.
Biophys Rev 6:75–87. https://doi.org/10.1007/s12551-013-0130-2
82. Cournia Z, Allen B, Sherman W (2017) Relative binding free energy calculations in drug
discovery: recent advances and practical considerations. J Chem Inf Model 57:2911–2937.
https://doi.org/10.1021/acs.jcim.7b00564
83. Huang SY, Zou X (2010) Advances and challenges in protein-ligand docking. Int J Mol Sci
11:3016–3034. https://doi.org/10.3390/ijms11083016
84. Maia EHB, Assis LC, de Oliveira TA, Da Silva AM, Taranto AG (2020) Structure-based
virtual screening: from classical to artificial intelligence. Front Chem 8:343. https://doi.org/
10.3389/fchem.2020.00343
85. Lin X, Li X, Lin X (2020) A review on applications of computational methods in drug
screening and design. Molecules 25:1375. https://doi.org/10.3390/molecules25061375
86. Amin SA, Banerjee S, Singh S, Qureshi IA, Gayen S, Jha T (2021) First structure-activity
relationship analysis of SARS-CoV-2 virus main protease (Mpro) inhibitors: an endeavor on
COVID-19 drug discovery. Mol Divers 25:1827–1838. https://doi.org/10.1007/s11030-020-
10166-3
2 In Silico Discovery of Class IIb HDAC Inhibitors: The State of Art 55

87. Scior T, Bender A, Tresadern G, Medina-Franco JL, Martínez-Mayorga K, Langer T, Cuanalo-

Contreras K, Agrafiotis DK (2012) Recognizing pitfalls in virtual screening: a critical review.
J Chem Inf Model 52:867–881. https://doi.org/10.1021/ci200528d
88. Kozikowski AP, Chen Y, Gaysin AM, Savoy DN, Billadeau DD, Kim KH (2008) Chem-
istry, biology, and QSAR studies of substituted biaryl hydroxamates and mercaptoac-
etamides as HDAC inhibitors nanomolar-potency inhibitors of pancreatic cancer cell growth.
ChemMedChem 3:487–501. https://doi.org/10.1002/cmdc.200700314
89. Tang H, Wang XS, Huang XP, Roth BL, Butler KV, Kozikowski AP, Jung M, Tropsha A
(2009) Novel inhibitors of human histone deacetylase (HDAC) identified by QSAR modeling
of known inhibitors, virtual screening, and experimental validation. J Chem Inf Model 49:461–
476. https://doi.org/10.1021/ci800366f
90. Zhao L, Xiang Y, Song J, Zhang ZA (2013) A novel two-step QSAR modeling work flow
to predict selectivity and activity of HDAC inhibitors. Bioorg Med Chem Lett 23:929–933.
https://doi.org/10.1016/j.bmcl.2012.12.067
91. Pham-The H, Casañola-Martin G, Diéguez-Santana K, Nguyen-Hai N, Ngoc NT, Vu-Duc
L, Le-Thi-Thu H (2017) Quantitative structure–activity relationship analysis and virtual
screening studies for identifying HDAC2 inhibitors from known HDAC bioactive chemical
libraries. SAR QSAR in Environ Res 28:199–220. https://doi.org/10.1080/1062936X.2017.
1294198
92. Zeb A, Park C, Son M, Rampogu S, Alam SI, Park SJ, Lee KW (2018) Investigation of non-
hydroxamate scaffolds against HDAC6 inhibition: a pharmacophore modeling, molecular
docking, and molecular dynamics simulation approach. J Bioinform Comput Biol 16:1840015.
https://doi.org/10.1142/S0219720018400152
93. Debnath S, Debnath T, Bhaumik S, Majumdar S, Kalle AM, Aparna V (2019) Discovery of
novel potential selective HDAC8 inhibitors by combine ligand-based, structure-based virtual
screening and in-vitro biological evaluation. Sci Rep 9:1–14. https://doi.org/10.1038/s41598-
019-53376-y
94. Sharma M, Jha P, Verma P, Chopra M (2019) Combined comparative molecular field analysis,
comparative molecular similarity indices analysis, molecular docking and molecular dynamics
studies of histone deacetylase 6 inhibitors. Chem Biol Drug Des 93:910–925. https://doi.org/
10.1111/cbdd.13488
95. Ruzic D, Petkovic M, Agbaba D, Ganesan A, Nikolic K (2019) Combined ligand and fragment-
based drug design of selective histone deacetylase-6 inhibitors. Mol Inform 38:1800083.
https://doi.org/10.1002/minf.201800083
96. Yan G, Li D, Zhong X, Liu G, Wang X, Lu Y, Qin F, Guo Y, Duan S, Li D (2021) Identification
of HDAC6 selective inhibitors: pharmacophore based virtual screening, molecular docking
and molecular dynamics simulation. J Biomol Struct Dyn 39:1928–1939. https://doi.org/10.
1080/07391102.2020.1743760
97. Uba AI, Yelekçi K (2019) Homology modeling of human histone deacetylase 10 and design
of potential selective inhibitors. J Biomol Struct Dyn 37:3627–3636. https://doi.org/10.1080/
07391102.2018.1521747
98. Uba AI, Yelekçi K (2020) Crystallographic structure versus homology model: a case study
of molecular dynamics simulation of human and zebrafish histone deacetylase 10. J Biomol
Struct Dyn 38:4397–4406. https://doi.org/10.1080/07391102.2019.1691658
99. Géraldy M, Morgen M, Sehr P, Steimbach RR, Moi D, Ridinger J, Oehme I, Witt O, Malz
M, Nogueira MS, Koch O (2019) Selective inhibition of histone deacetylase 10: hydrogen
bonding to the gatekeeper residue is implicated. J Med Chem 62:4426–4443. https://doi.org/
10.1021/acs.jmedchem.8b01936
100. Ukey S, Choudhury C, Sharma P (2021) Identification of unique subtype-specific interaction
features in Class II zinc-dependent HDAC subtype binding pockets: A computational study.
J Biosci 46:1–22. https://doi.org/10.1007/s12038-021-00197-9
Chapter 3
Role of Computational Modeling in Drug
Discovery for Alzheimer’s Disease

Mange Ram Yadav, Prashant R. Murumkar, Rahul Barot, Rasana Yadav,

Karan Joshi, and Monica Chauhan

Abstract Researchers are striving hard for the last two decades to discover new
therapeutically effective molecules for the treatment of Alzheimer’s disease (AD).
Unfortunately, the exact etiology of Alzheimer’s disease is not yet fully known,
which is proving to be the main hurdle for the discovery of drugs to treat the
disease. Several factors are involved in the etiology of AD, like oxidative stress,
low levels of acetylcholine (ACh), β-amyloid aggregation, and tau protein phos-
phorylation in the brain. But unfortunately, no single drug has proved clinically
effective till today to prevent or stop progression of the disease. The existing drugs
simply improve the worsening clinical symptoms of AD and help in delaying the
process of progression of the disease to a fully blown state. Currently, available
drugs in the market for the treatment of AD include donepezil, galantamine and
rivastigmine, the three acetylcholinesterase inhibitors (AChEIs), and memantine an
N-methyl-D-aspartate receptor (NMDAR) antagonist. These drugs are used mainly
to alleviate mild cognitive impairment (MCI) providing temporary relief from the
symptoms. This chapter discusses about the application of various computational
tools for compounds containing different heterocyclic moieties like quinoline, pyri-
dine, pyrimidine, coumarine, chromane, indole, etc., which could serve as potential
leads to design potent novel anti-Alzheimer’s agents.

Keywords Alzheimer’s disease · QSAR · Molecular modeling · Docking ·

Molecular dynamics · Drug research

M. R. Yadav (B)
Centre of Research for Development, Parul University, Waghodia Road, Vadodara,
Gujarat 391760, India
e-mail: [email protected]
P. R. Murumkar · R. Barot · R. Yadav · K. Joshi · M. Chauhan
Faculty of Pharmacy, The Maharaja Sayajirao University of Baroda, Vadodara, Gujarat 390001,
India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 57

S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_3
58 M. R. Yadav et al.

3.1 Introduction

Alzheimer’s disease (AD) is a progressive, incurable, and debilitating neurodegen-

erative disease that eventually leads to a state of dementia. Currently, around 50
million people are affected by dementia worldwide, and by 2050 that number may
rise to over 150 million making AD the seventh major cause of deaths [1, 2]. The
early signs of Alzheimer’s include the presence of β-amyloid (Aβ) plaques and
neurofibrillary tangles (NFTs) in the central nervous system (CNS). Two main types
of Alzheimer’s disease exist, the early-onset familial Alzheimer’s (FAD), and the
sporadic Alzheimer’s (SAD) which has late onset. FAD is hereditary and typically
begins early in life. It tends to affect the first-degree relatives of the patients. It is
believed to be caused by mutations in three genes involved in the expression of
amyloid precursor protein (APP), presenilin 1 (PSEN1), and presenilin 2 (PSEN2).
SAD, on the other hand, is considered to be caused by genetic and environment factors
and generally begins in later part of life. Mainly, Eε4 (ε4 allele of ApoE gene), a
gene apolipoprotein, is responsible for SAD, while FAD has a strong genetic link
to chromosomes 1, 14, 19, and 21. The progression of Alzheimer’s is divided into
three stages: preclinical AD, AD with mild cognitive impairment (MCI), and AD
causing dementia. The dementia phase of Alzheimer’s is further divided into three
levels of mild, moderate, and severe, which indicate the manifestation of the degree
of symptoms affecting a person’s ability to perform daily activities [3, 4].
Low levels of the neurotransmitter acetylcholine (ACh) due to high levels of
cholinesterase (ChE) enzymes and hyperactivity of N-methyl-D-aspartate (NMDA)
glutamatae receptors in the brain are the two main biochemical reasons for cognitive
impairment in AD patients. Increased free radicals and an increased influx of ions
caused by hyperactivity of glutamate NMDA receptors in the brain prove detrimental
to the neurons [3].
Hypothesis Behind the Development of AD
Several hypotheses have been proposed by the researchers for explaining the genesis
of AD in human beings over the years. These hypotheses are mainly classified as
follows:

3.1.1 The Cholinergic Hypothesis

Acetyl choline, a neurotransmitter plays important role in cognitive processes. As per

the cholinergic hypothesis, synaptic acetylcholine deficit and impairment of cholin-
ergic neurons lead to the indisputable causes of AD. These lead to decrease in memory
and learning abilities and neurobehavioral actions such as agitation and aggres-
sion. AChE and BuChE are responsible for decreased levels of ACh in the brain as
they rapidly hydrolyze the neurotransmitter into acetate and choline [5]. In AD, the
levels of both AChE and BuChE are increased, which cause decrease in the levels of
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 59

ACh. Additionally, studies have shown that both of the cholinesterases interact with
amyloid-β (Aβ) to form aggregates that cause greater neurotoxicity than the free Aβ.
The AChE-Aβ complex causes an uncontrolled Ca2+ influx into the neurons’ interior
chamber that leads to dysfunction in mitochondria and loss of membrane potential.
Thus, inhibition of the cholinesterase enzymes remains the most common approach
for AD treatment [6].

3.1.2 The Amyloid Hypothesis

β-amyloid has emerged as the most focused target for the treatment of AD because
synaptic plasticity is significantly disrupted by Aβ that induces the formation of reac-
tive oxygen species (ROS), which are responsible for the disability of mitochondrial
function and neuro-inflammation [6]. Studies suggest that Aβ is involved in many
processes like formation of extracellular plaques and disruption of neuronal cell
structure, leading to the aggregation and accumulation of Aβ fibrils in the neurons
[7, 8].

3.1.3 Tau Protein Hypothesis

Tau protein is another crucial protein which is responsible for the genesis of AD.
Recent studies have shown that in sporadic AD, the tau hypothesis may be the initiator
of the events. Post-translational hyper-phosphorylation of tau protein leads to forma-
tion of aggregates similar to Aβ, known as neurofibrillary tangles (NFTs). However,
it needs to be clarified that Aβ and NFTs are not exclusively correlated to AD progres-
sion, because tau hyper-phosphorylation and tau production abnormalities have also
been observed in other types of neurodegenerative disorders.
For the treatment of Alzheimer’s disease, three drugs are acetylcholinesterase
inhibitors, i.e., donepezil, galantamine, and rivastigmine, while memantine is an
N-methyl-D-aspartate (NMDA) receptor antagonist. Cholinesterase inhibitors are
advocated in mild to moderate cases of Alzheimer’s disease while in moderate to
severe cases, memantine is recommended. Combination of donepezil and memantine
has given better results than the use of either of the drugs all alone. A new drug,
Aducanumab, developed by Biogen, is the first human monoclonal antibody to be
approved that targets Aβ directly. It binds to both Aβ fibrils as well as the soluble
oligomers [9, 10].
60 M. R. Yadav et al.

O
H3CO
O CH3
N H3CO
H3C N O CH3
CH3 CH3 N

(Rivastigmine) (Donepezil)
OCH3
H
O NH2

HO H 3C
N CH3
CH3

(Galantamine) (Memantine)

Treatment Strategies for AD

Multi-target drugs (MTDs) approach is the buzzword used widely these days to treat
multifactorial diseases like AD. MTDs are defined as molecules in which two or
more pharmacophoric features are incorporated in a single molecular entity. This
single molecule can target different biological targets. Development of MTDs is
considered to be a promising strategy, but any MTD has yet to reach the clinician’s
desk and see the light of the day for the treatment of AD. More than 140 MTDs
eliciting different mechanism of action are currently in clinical trials in different
phases for AD treatment. Currently in the AD interventional studies, there are 111
new chemical entities in clinical trials either in phase 2 or phase 3 and 29 in phase
3. In order to reverse the progression of AD, a majority of the MTDs are in different
phases of clinical trials target neuro-inflammation, oxidative stress, mitochondrial
dysfunction, metal dysregulation, tau proteins, amyloids (both AChE and BACE), or
some miscellaneous targets [11–13] (Fig. 3.1).

3.2 Role of Computational Studies in the Designing

of Anti-Alzheimer’s Agents

In this chapter, we have endeavored to discuss about the role of various computa-
tional techniques for the designing of new anti-AD agents. Rational drug design and
discovery is highly dependent on computer-aided drug design techniques. There are
two important drug designing techniques that are mainly used in computer-aided
drug design, i.e., ligand and structure-based drug design techniques. Ligand-based
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 61

Fig. 3.1 Various targets

exploited for developing
anti-AD agents (reused with
permission source Fig. 1,
Oset-Gasque and
Marco-Contelles [13])

designing technique is also called as indirect drug designing technique wherein only
the structures of the ligand molecules are considered [14], whereas the structure-
based drug design or direct drug designing technique additionally requires the three
dimensional structure of the target enzyme/protein [15, 16]. In our own lab, we have
used CADD techniques for designing of TACE inhibitors [17–27] and anti-AD agents
[28–37].
This book chapter describes some recent applications of various computational
techniques to discover or design new potential anti-Alzheimer’s agents, by applying
different approaches like molecular docking, quantitative structure–activity relation-
ship (QSAR), and molecular dynamics simulations. Since a huge quantum of liter-
ature related to development of anti-AD agents is available in the public domain,
we have covered the literature reported from the year 2018 onward in this chapter,
related to some potent small molecules having activity in nanomolar range against
the main AD targets like AChE, BuChE, PDE, BACE1, amyloid β, GSK-3β, and also
having antioxidant and metal chelating properties. Chemical moieties falling under
different chemical classes are discussed in the following sections.

3.2.1 Tacrine-Based Scaffolds as Anti-AD Agents

Tacrine was the first US-FDA approved drug made available for the treatment of mild
to moderate cases of AD, but later on, it was withdrawn due to its severe side effects
like gastrointestinal discomfort and hepatotoxicity. But, some features of tacrine
like decrease in Aβ peptide-induced apoptosis in cortical neurons were observed
to be positive for the treatment of AD. Based on these observations, Przybyłowska
et al. designed novel N- and O-phosphorylated tacrine derivatives [38]. Among the
compounds of the reported series, compound (1) exhibited the highest inhibitory
activity with an IC50 value of 6.11 nM against AChE, while compound (2) offered
an IC50 of 1.97 nM against BuChE, proving compound (1) and compound (2) to be
6 and 12-fold more potent than tacrine, respectively.
62 M. R. Yadav et al.

O
OR
P
NH2 HN n N OR
H

N N
Tacrine
1. n = 8, R = Et
2. n = 12, R = Ph
3, n = 6, R = Ph

Molecular modeling studies revealed that all the compounds offered satisfactory
free binding energies, for AChE in the range of −9.2 kcal/M and for BuChE −7.8 to
−10.92 kcal/M compared to the reference compound tacrine which yielded binding
energies of −9.0 kcal/M for AChE and −8.2 kcal/M for BuChE. Compound (3)
proved to be the best fitting molecule for inhibiting the catalytic activity in the cavity
of AChE and BuChE with docking energies of −12.5 and −10.7 kcal/M, respectively.
In the CAS region of both of the enzymes, aromatic ring of tacrine showed π –π
stacking interaction with Phe330 and Trp84 (AChE), and Trp82 (BuChE), and H-
bonding between the nitrogen atom of tacrine and His440 and His438 of AChE and
BuChE, respectively. Additionally, the phosphoramide group formed H-bond with
the Tyr121 and Thr110 in the mid-gorge regions of AChE and BuChE, respectively,
while non-covalent interactions were observed with Trp279 of AChE.
In the pathology of AD, hyper-phosphorylation of tau protein plays a signifi-
cant role resulting into microtubule polymerization and NFT generation that lead to
necrosis and degeneration of the neuronal cells, leading to progressive deterioration
of the patient’s condition. Glycogen synthase kinase 3 beta (GSK-3β) catalyzes
the disordered phosphorylation of tau protein. Therefore, tau protein signaling
pathway involving GSK-3β has become a vital target for designing of new anti-AD
compounds. Based on this philosophy, Yao et al. reported novel tacrine-pyrimidone
hybrids as potent dual GSK-3β and AChE inhibitors as anti-AD agents [39]. In the
reported series of compounds, 4 exhibited excellent dual GSK-3 and AChE inhibition
activities with IC50 values of 89.3 and 51.1 nM, respectively.
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 63

O
H3C
N
HN 5 N N
H
N
F
Cl N
(4)
As per the docking studies, compound (4) was perfectly fitting in the catalytic
anionic site (CAS). The tacrine fragment was positioned in the interior gorge,
while the pyrimidone fragment was located in the exterior gorge of the peripheral
anionic site (PAS) of AChE. Five non-bonding interactions were observed between
compound (4) and the binding gorge of the enzyme. H-bond between chloro group of
compound (4) and oxygen atom of Tyr72 (3.5 Å) drives the tacrine unit to bind in the
reverse way, costing one π–π interaction with Trp86, but additionally making two
π–π stacking interactions with Trp286, and H-bonding with Ser293 with a distance
of 2.2 Å (Fig. 3.2).
Pyrimidone fragment of compound (4) intruded into GSK-3β protein’s interior
part of the binding pocket, while the tacrine unit got stuck to the exterior portion of
the active pocket (Fig. 3.3a, b). The hybrid molecule (4) made a π-cation interaction
(4.5 Å) with Arg141, and two H-bonds with Val-135 and Lys-85 with distances of
2.6 and 2.7 Å, respectively, on the opposite side of the peripheral groove (Fig. 3.3c).
Compound (4) was fitting into the binding pockets of GSK-3β and AChE and showed
good affinity for both the targets by interacting with several amino acid residues
making secondary bonds through the cooperation of the tacrine unit, pyrimidone

Fig. 3.2 Binding of compound (4) with AChE. a, b are shown using the PDB code 4EY7. The
position of compound (4) in brown is depicted within the pocket of AChE, and the interactions are
shown between compound (4) and the amino acid residues. In c, the 2D binding mode of compound
(4) with AChE is depicted, hydrogen bonds shown in green, the π–π stacking interaction, shown
in purple, and halogen bond, shown in cyan (reused with permission source Fig. 3, Yao et al. [39])
64 M. R. Yadav et al.

Fig. 3.3 Interaction of compound (4) with GSK-3β, a, b display the location of compound (4) in
the active site of the protein, and c shows a 2D representation via hydrogen bonds and π-cation
bond (reused with permission source Fig. 3, Yao et al. [39])

moiety, and the alkylamine linker catapulting it an excellent dual GSK-3β and AChE
inhibitor.
Furthermore, Ozten et al. reported carbamate derivatives of tacrine as potent
cholinesterase inhibitors. Compound (5) showed the best inhibitory activity against
both the enzymes BuChE and AChE with IC50 values of 16.96 and 22.15 nM,
respectively [40].

H
O 2N N O
NH
O

N
(5)

In the modeling research, it has been noticed that compound (5) interacts with
AChE and BuChE enzymes through π-cation and π–π stacking interactions with
the phenyl ring, in a similar manner to tacrine and its derivatives. Compound (5)
forms strong hydrogen bonds to AChE, with residues His478 in the catalytic site,
Glu233 in the choline binding region, and Gly153 in the acyl-binding region and
also has a π-cation interaction in the acyl-binding site with Phe369 through its -
NO2 group. Additionally, compound (5) interacts with water molecules and forms
stable bridges with Tyr360 in the BuChE binding cavity and Glu225 in the choline
binding site. The residues from the acyl-binding region also play a role in stabilizing
compound (5) in the BuChE binding site through hydrogen bonding with Gly143 and
Gly145. Although some non-covalent interactions were seen to increase the affinity
of compound (5) temporarily for AChE and BuChE, but these were not sustained
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 65

during the entire MD simulation. The MD study suggests that compound (5) forms
covalent bonds with residues Ser226 and Ser234 for AChE and BuChE, respectively.
Chufarova et al. have reported hybrid molecules of tacrine and acridine as potential
anti-acetylcholinesterase and anti-butyrylcholinesterase enzymes [41]. Compound
(6) was found to be the most potent inhibitor having IC50 values of 1.7 and 7.6 pM
against BuChE and AChE enzymes, respectively. Compound (6) was found to be
having better activity than the standard tacrine without any significant cytotoxic
effect.

O NH

5 NH

(6)
Modeling studies revealed significant interactions by the potent compounds in
the active site of the enzyme when compared to tacrine. Compound (6) was found
to be having dual-binding mode (bound to PAS and CAS both). It was oriented in
the opposite direction of tacrine making π–π stacking with residues Tyr341 and
Trp286 along with the two T-shaped aromatic interactions with residues Tyr124
and Tyr72 in the PAS site of AChE. Similarly in CAS site, it showed T-shaped
interaction with residue Tyr124 and π–π stacking with Trp86 residue. Whereas, in
BuChE, the compound showed very limited interactions. Tacrine moiety interacted
with residue Tyr332 and acridine moiety interacted with Phe329 residue by two π-
stacking interactions. There were no T-shaped interactions observed in the BuChE
site.
Li et al. reported a novel series of tacrine-phenolic acid and tacrine-phenolic acid-
ligustrazine, dihybrids, and trihybrids, respectively, as acetylcholinesterase inhibitors
[42]. Among them, compound (7) displayed potent activity as acetylcholinesterase
inhibitor with IC50 value of 3.9 nM; hAChE IC50 value of 65.2 nM, along with the
blocking of β-amyloid self-aggregation effect at 20 μM.
66 M. R. Yadav et al.

O
OCH3
HN N
H
OH
OCH3
N

(7)
In order to see the molecular interactions, the most potent compound (7)
was subjected to docking studies in the active site of the enzyme cholinesterase.
Docking studies using TcAChE (PDB ID 3CKM) revealed that tacrine component
of compound (7) interacted within the CAS site with phenyl ring of Phe330 residue
and indole moiety of Trp84 via aromatic π–π stacking, whereas the sinapic acid
component interacted via hydrophobic interactions with residues Ser286 and Asp285
within the PAS site. Similarly, using AChE (PDB ID: 4EY7) active site, these studies
revealed that sinapic acid component made aromatic π–π stacking interaction with
Trp286 residue in the PAS site, and the tacrine moiety was involved through the
hydrophobic interactions with residue Tyr124 that offered stacking interaction with
indole ring of Trp86. These docking interactions in the CAS and PAS sites elicited
the active binding of compound (7) with TcAChE and hAChE. Compound (7) has
shown π–π stacking interaction between Trp82 residue and tacrine moiety in the
active site of hBuChE (PDB ID 1P0I) and hydrogen bonding interactions between
Ser198 residue and hydroxyl group.
Derabli et al. disclosed poly-functionalized tacrine-derived compounds via
Friedlander reaction and evaluated them for cholinesterase (BuChE and AChE)
inhibitory activity [43]. Among the reported derivatives, compound (8) containing 4-
chlorophenyl and S-ethyl moieties yielded the most potent compounds against AChE
(IC50 value of 4.32 μM), while compound (9) elicited higher potency against BuChE
with IC50 value of 2.74 μM.

OCH3
Cl

NH2
NH2
N
N
H3C S N N
H 3C S N N

(8) (9)

Molecular modeling studies were performed to find out the binding interactions
of both of the compounds (8 and 9), in the active sites of BuChE (PDB ID: 5dyw)
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 67

and AChE (PDB ID: 4EY7). Biological evaluation confirmed that compound (8)
was more potent against AChE than compound (9), which was more potent than
compound (8) against BuChE. In the AChE active site, compound (8) exhibited
hydrophobic interactions with Trp8 residue and hydrogen interactions with Tyr124,
as the heterotacrine moiety of the compound (8) was much better oriented to offer a
good binding affinity. Whereas with BuChE, compound (9) showed pi-interactions
with phenyl ring of Tyr332 residue and hydrogen bond interaction with Asp70
residue.
Zhu et al. synthesized a series of tacrine-ferulic acid hybrids as multi-target-
directed ligands [44]. Among the reported series of compounds, compound (10)
displayed promising activity as BuChE inhibitor with IC50 value of 101.40 nM, and
as AChE inhibitor with IC50 value of 37.02 nM. The compound (10) also exhibited
potent amyloid β-protein self-aggregation inhibitory activity.

CH3

O
CH3
H
N
HN OCH3
O

(10)

To investigate the intermolecular interactions of compound (10) in the huAChE

(PDB ID: 4EY7) and huBuChE (PDB ID: 4TPK) active sites, molecular docking
studies were performed. From these studies, it was revealed that in huAChE 1,2,3,4-
tetrahydroacridine moiety of compound (10) made π–π stacking interactions with
residue Trp86 and hydrogen bonding with residue Tyr337 in the CAS binding site of
the enzyme. Whereas the ferulic acid phenyl ring interacted with Trp286 and Tyr341
residues via π–π stacking interactions within the PAS site. This phenyl ring also
interacted with residues Leu289, Ser293, and Tyr124 via van der Waals interactions
(Fig. 3.4).
Wiȩckowska et al. have carried out the synthesis of multi-target-directed ligands
and reported a compound (11) displaying potent activity as an antagonist of 5-HT6
receptor and non-competitive inhibitor of cholinesterase (IC50 , hAChE = 14 nM,
IC50 , eqBuChE = 22 nM) and amyloid-β anti-aggregation activity (IC50 = 1.27 uM)
[45]. From the docking studies, it has also been revealed that in the active site of
the enzyme, the short chains (2–3 carbon atom linkers) and long chains (6–8 carbon
atom linkers) exhibit distinct binding modes. Short chain derivatives of the tacrine
fragment interact with Phe330 and Trp84 residues of the catalytic binding site of
AChE, whereas π–π stacking interaction is formed with the 1-(3-(benzyloxy)-2-
methylphenyl)piperazine fragment and methylphenyl or indole fragment of Trp279
residue of the peripheral site of AChE, while π-cation interaction takes place between
68 M. R. Yadav et al.

Fig. 3.4 Binding mode of compound (10) with huAChE. Binding interactions are indicated in
different colors as dashed lines. Red: π –π T-shaped, purple: π –π stacking and green color: hydrogen
bond interaction (reused with permission source Fig. 1, Zhu et al. [44])

Tyr334 residue and piperazine. Similarly, long chain compounds show same interac-
tions but piperazine interacts with Phe331 instead of Tyr334 and forms cation-l inter-
action, whereas in short chain compounds, tacrine component forms π–π stacking
interaction with Tyr334, and rest of the molecule exists in perpendicular positioning
the catalytic site. Compounds with 5 carbon linkers were showing poor binding
affinity as they were unable to make interactions with residues Trp279, Tyr334 or
Phe331.

N
H S
N
HN 4 N CH3 HN N
4
N O CH3

N N

(11) (12) X = Cl; (13) X = F

Makhaeva et al. reported a series of conjugates of tacrine-1,2,4-thiadiazole deriva-

tives using quantum mechanical characterization followed by molecular docking
which revealed exceptional binding affinity of these compounds for the PAS and
CAS active sites of BuChE and AChE both. The in vitro assay results justified the
in silico predictions for these compounds. The compounds displaced propidium dye
from PAS, indicating that the compounds were blocking AChE-induced β-amyloid
aggregation [46]. Molecular docking studies were performed using PDB ID: 4EY7
for AChE and 1POI for BuChE using AutoDock 4.2.6. Molecular docking studies
predicted dual binding in both of the active sites of AChE in which the tacrine
part formed π-cation interactions and π–π stacking with Trp86, and ionic inter-
actions with Glu2020 along with hydrogen bonding stabilized these interactions.
The pentamine in the structures (12, 13) additionally produced ionic and π-cation
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 69

interactions between the amino acid residues in the gorge, mainly with Tyr341 and
the charged amino group of these compounds. Molecular docking studies predicted
that the substituents on the aromatic group on the thiadiazole moiety did not produce
substantial anti-AChE effect because of small gorge cavity area in the target enzyme,
which was not the case with BuChE, as the compounds could bind tightly in the
active site in folded conformation. Compound (12) containing pentamine fragment
achieved better results compared to pentamine conjugates of compound (13). The
compounds (12 & 13) offered in vitro IC50 values of 0.00380 ± 0.00029 μg/mol and
0.0725 ± 0.0016 μg/mol, respectively, as anti-BuChE agents, when compared with
the standard drug tacrine (anti-BuChE IC50 value of 0.0295 ± 0.0020 μg/mol).

3.2.2 Indole-Based Anti-AD Agents

In the year 2021, a series of sulfonamide group containing dihydropyranoindoles

were reported by Sarfaraz S et al. along with their in vitro AChE activity. All the
synthesized compounds showed AChE (IC50 = 0.41–8.79 μM) and BuChE (IC50
= 1.17–30.17 μM) inhibitory activity [47]. Among the compounds of the reported
series, compound (14) showed the highest inhibitory activity for BuChE (IC50 =
1.17 μM) as well as AChE (IC50 = 0.41 μM) when compared with galantamine and
rivastigmine as the standard drugs. Moreover, the enzyme kinetics revealed that the
compound (14) possessed mixed type of inhibition, and it was bound to catalytic sites
(CAS) as well as to the peripheral anionic sites (PAS) of both the enzymes. Compound
(14) was docked in the active site of AChE (Fig. 3.5) to understand its binding mode.
The modeling studies revealed that it interacted with active sites of both the enzymes
BuChE and AChE. Compound (14) showed multiple H-bonding interactions with
Asp72 residue of PAS and through Ser200 residue with CAS and acyl-binding pocket
with Phe288 residue. Two π–π interactions were observed with Trp334 and Tyr279
residues in the PAS while one π–π interaction was formed at anionic sub-site with
Phe330 residue. For comparison purposes, galantamine and rivastigmine were also
docked into the active site of AChE wherein interactions similar to those observed
with compound (14) were seen. The orientations of compound (14) galantamine
and rivastigmine with AChE are shown in Fig. 3.5. In case of BuChE, compound
(14) formed two H-bonding interactions with Ser287 residue in the acyl pocket and
Glu197 residue in the PAS. It also displayed π–π stacking interactions with His438
of CAS, Trp82 of PAS and Phe329 of the anionic sub-site. These results indicate
mixed type of inhibition by compound (14) of BuChE. Compound (14) exactly fits
into BuChE binding pocket with similar interactions as seen in the case of standard
drugs.
70 M. R. Yadav et al.

Fig. 3.5 Docking interaction of a compound (14), b rivastigmine, and c galantamine within the
active site of AChE (reused with permission source Fig. 6, Shaikh et al. [47])
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 71

OMe
MeO OMe

N
O O
S O NH2
N

Me

(14)

He et al. found that melatonin-ferulic acid derivatives can act as selective inhibitors
of HDAC6, an enzyme involved in memory, neurodevelopment, and cognitive
processes. Inhibitors of HDACs have been recognized as a potential treatment for
AD type neurodegenerative disorders, wherein HDAC6 plays an important role
in deacetylation of tau protein that leads to aggregation of phosphorylated tau
protein into neurofibrillary tangles. The study showed that the melatonin-ferulic
acid derivatives have immunomodulatory and neuroprotective effects [48].

OH
H3CO

H3CO N
O

N
H
O
HN
OH
(15)

Compound (15) showed selective inhibition of HDAC6 among the reported series
of compounds, with an IC50 value of 30.7 nM, and it exhibited more than 25-fold
higher selectivity than the other reported compounds from the series. Moreover,
compound (15) was comparable to ferulic acid in demonstrating DPPH radical scav-
enging activity. To identify putative binding modes of compounds (15) with HDAC6,
computational docking studies were performed using PDB: 6DVM. For the capping
groups, two binding modes could be envisaged, i.e., the binding groups could be
flanked by loops L1 and L2 above the pocket. In one of the binding modes, the
indole moiety showed aromatic interactions with His463, whereas van der Waals
interactions were seen between hydroxymethoxyphenyl ring and Leu712 residue.
72 M. R. Yadav et al.

Fig. 3.6 Interaction of compound (15) with HDAC6 (PDB: 6DVM) (reused with permission
source Fig. 2, He et al. [48])

As far as the switched orientation is concerned, it was found less favorable as the
indole ring was placed above Leu712, and the -OH group of the ferulic acid-based
moiety approached Asp460 for H-bond interaction in the docking studies. On the
basis of these observations, it was assumed that the likely mode of binding of the
compound (15) was binding of both of the capping groups into the L1 loop pocket
of HDAC6, offering selectivity to the compound for the enzyme HDAC6 (Fig. 3.6).
Ghamari et al. disclosed a series of H3R antagonist and anti-cholinesterase
dual acting compounds. They have identified compound (16), as the lead molecule
using computational studies [49]. The molecular docking studies showed its effec-
tive binding interactions within the active sites of the enzymes. For cholinesterase
inhibitory activity, PDB IDs selected for AChE were 4EY7 and for BuChEit were
4TPK. GOLD software was used to perform the docking studies. In modeling
studies, it was seen that compound (16) exhibited one l–l stacking with the residue
Trp286, two hydrophobic interactions with Tyr337, Tyr341, Trp86, and His447,
and one hydrogen bonding with the amino acid Tyr124 within the active site of
AChE; whereas in BuChE active site, this compound has shown hydrophobic inter-
actions with residues Trp82, Tyr128, and Glu197 forming a hydrophobic pocket
and a hydrogen bonding interaction with residue Tyr332. It was revealed from the
in vitro studies that the compounds of the series possessed potent H3R antagonist
and cholinesterase inhibitory activities. Compound (16) showed H3R antagonistic
activity with EC50 value of 0.73 nM, AChE inhibitory activity with IC50 value of
9.09 μM, and BuChE inhibitory activity with IC5 0 value of 21.1 μM.
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 73

N OH
H3C Cl H3CO H
N CH3 N
N N
N
O H O

(16) (17)

Lee et al. disclosed a series of 5-aroylindolyl-substituted hydroxamic acids.

Among them, compounds (17) displayed the most potent inhibitory activity against
histone deacetylase 6 (HDAC6) along with the reduction of tau protein phosphory-
lation and aggregation [50]. The phosphorylated tau protein was transferred to the
Hsp70/CHIP complex resulting into the ubiquitination of the phosphorylated tau
protein. Its inhibitory activity also results into increased levels of acetylated Hsp90
that affects the binding of Hsp90 to HDAC6.
Docking studies were also undertaken along with the in vitro experimentation, to
elucidate the selectivity of the synthesized compounds. Different crystal structures
of histone deacetylase like HDAC1 (PDB ID: 5ICN), HDAC2 (PDB ID: 5IX0),
HDAC3 (PDB ID: 4A69), HDAC6 (PDB ID: 5EDU), and HDAC8 (PDB ID: 1W22)
were used for docking of compound (17) in order to determine its selectivity. It was
revealed that compound (17) was bound in a specific pocket of protein (5EDU) which
is not present in any other histone deacetylase and its anisole moiety played a very
important role in its selectivity as it interacted with the residues N494, D496, and
W497 to form van der Waals interactions and with residue N494 to forms hydrogen
bonding (Fig. 3.7).
Glycogen synthase kinase 3β (GSK-3β) has been identified as a novel target
to treat Alzheimer’s disease. A series of GSK-3β inhibitors has been explored by
Lozinskaya et al., A compound (18) bearing 3-arylidene-2-oxindole scaffold has been
reported as a novel GSK-3β inhibitor with an IC50 value of 4.19 nM for the enzyme

Fig. 3.7 Interaction of compound (17) with 5EDU (reused with permission source Fig. 4a, Lee
et al. [50])
74 M. R. Yadav et al.

[51]. Docking studies were performed using Autodock Vina 1.1.2 using the protein
structure of GSK-3β (PDB ID: 3SD0 and 4J1R). The oxindole moiety of compound
(18) occupied the hydrophobic cleft of the allosteric site of the ATP-binding site
of GSK-3β in the docking studies. It exhibited tight alkyl-π interaction with the
lipophilic side chains of Ala83, Val110, Leu132, Leu188, and Cys199 for both of
the geometric isomers. A characteristic feature of the allosteric GSK-3β inhibitors,
i.e., hydrogen bond interaction between NH and Asp133 was also observed.

O
N
H
(18)

3.2.3 Pyridine and Pyrimidine-Based Scaffolds as Anti-AD

Agents

Shoaib et al. designed and synthesized a novel series of phenylsulfonyl-pyrimidine

carboxylate MTDLs for the treatment of AD. Compound (19) displayed potent
inhibitory activity against AChE having IC50 value of 47.33 ± 0.02 nM among
the reported series of compounds. Compound (19) displayed non-competitive inhi-
bition of AChE in the enzyme kinetics study with Ki value of 8 nM. AChE-
induced Aβ1-42 aggregation was also inhibited by the compound at a concentration
of 10 μM in the thioflavin T-assay [52]. All the compounds of the series displayed
high stability in CAS site of AChE enzyme with binding free energy ranging from
−12.38 to −9.69 kcal/M through π –π interactions and H-bondings in the

OCH3 O O
H3CO S Cl
N
N N NH
H3C O N O CH3
O CH3

(19)
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 75

docking studies. The phenyl ring of the 3,4-dimethoxyphenyl component of

compound (19) displayed π –π stacking interactions (distance 3.8 Å) with Trp279
residue in the PAS sub-site (Fig. 3.8). Indole ring system of Trp84 residue
was involved in sandwich-type π –π stacking with the aromatic ring of o-
chlorophenylcetamido group (distance 4.1 Å) of compound (19) in the CAS sub-
site of the enzyme. Methoxy group of compound (19) offered strong H-bonding
with Arg289 residue at a distance of 1.8 Å. A hydrogen bond (3.1 Å) was formed
between pyrimidine ring nitrogen and Tyr121 residue. Additionally, methoxyl group
of compound (19) formed H-bonding with Ser286 (3.2 Å). Carbonyl group was also
involved in forming H-bond with Ser124 and Gly123 at distances of 3.3 and 2.7 Å,
respectively. Based on the docking results, it could be concluded that the selected
compound possessed dual-binding inhibitory activity for AChE by targeting PAS and
CAS sub-sites. Similarly, using AutoDock software, docking studies for compound
(19) were further extended on hBuChE employing the crystal structure of hBuChE
(PDB ID: 4TPK) as shown in Fig. 3.9. It displayed better interactions (binding free
energy −8.7 kcal/M) with the enzyme. In the CAS sub-site, Trp82 residue of the
enzyme interacted with the aromatic ring through π –π stacking interaction. Ser70
fragment of the enzyme interacted with the methoxyl group through H-bonding,
while the NH-group of acetamido function made the strongest bond with Glu115
fragment of the enzyme. The docking results clearly indicate that these pyrimidine-
phenylsulfonyl derivatives target both the sub-sites, i.e., PAS and CAS of AChE as
well as of BuChE.
Zhang et al. reported a novel series of pyrazole-pyrimidinone derivatives as
PDE9A inhibitors having additional antioxidant activity, as anti-AD agents in the
year 2018. Out of 14 compounds, 12 compounds showed IC50 values below 200 nM
along with good antioxidant activity. Compound (20) among the reported series of
compounds offered an IC50 value of 56 nM, the lowest one making it the most
potent compound of the series [53]. Compound (20) proved to be safe toward human
neuroblastoma (SH-SY5Y) cells in the toxicity studies. From the structure–activity

Fig. 3.8 Docking interactions of compound (19) within the active site of AChE (PDB code: 1EVE)
(reused with permission source Fig. 3, Manzoor et al. [52])
76 M. R. Yadav et al.

Fig. 3.9 Docking interactions of compound (19) within the active site of BuChE (PDB code: 4TPK)
(Source Fig. 4, Manzoor et al. [52])

relationship and molecular docking studies, it was revealed that for good selectivity
for PDE9A, the compounds needed to show interactions with Tyr424 as it is unique
in PDE9. Along with this, the compounds also needed to interact with Gln453 and
Phe456.

HN
H N
N N
HO N
O CH3

(20)

In 2018, Ghobadian et al. reported a series of some carbazole-benzyl-pyridine-

based BChE inhibitors [54]. From the in vitro activity, it was found that all these
derivatives displayed potent and selective anti-BuChE activity having strong BuChE
inhibition in the range of 0.073 to 1.6 μM. Compound (21) exhibited mixed type of
inhibition with the highest selectivity and potency with an IC50 value of 0.073 μM.
Docking studies performed on the compound showed that it was bound to the catalytic
site, peripheral site, the choline site, and the acyl pocket of BuChE. It was bound to the
enzyme with a free binding energy of −8.68 kcal/M. BuChE choline binding site was
fully occupied by the planar part of the carbazole moiety. No violation of Lipinski’s
rule of five was observed during the screening of these derivatives. Additionally,
the compound (21) was found to be neuroprotective at 10 μM. It was also observed
that the compound could inhibit self-induced and AChE-induced aggregations at
concentrations of 10 and 100 μM, respectively.
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 77

Cl N N

N O N
N N N
N
.Cl
N

(21) (22)

A hybrid series of triazolopyrimidine and pyrimidine derivatives was reported

by Kumar et al. for the treatment of Alzheimer’s disease as potent acetyl-
cholinesterase inhibitors [55]. 2-(4-(6-(Quinolin-8-yloxy)pyrimidin-4-yl)piperazin-
1-yl)nicotinonitrile (22) with an IC50 value of 36 nM was found to be the most
potent compound of the series. Using the crystal structures of TcAChE and rhAChE,
docking studies were undertaken, in order to see the interactions between the hybrid
molecules and the binding residues in the active site of AChE. Docking studies
revealed that compound (22) interacted at three different sites of rhAChE, namely
PAS, CAS, and catalytic site in its active binding domain. The cyanopyridine and
8-oxyquinoline moieties of compound (22) interacted with the active site residues
Trp86 and Trp286 of the enzyme via π–π stacking.

H N
N
N
H
N N O
N NH
N N
HN
O

(23)

2-(Piperazin-1-yl)-N-(1H-pyrazolo[3,4-b]pyridin-3-yl)acetamide derivatives
were reported by Tarana Umar et al. as potent AChE and amyloid-β aggregation
inhibitors. Docking studies revealed strong interactions of the designed ligands with
the amino acid residues of the active site of the protein which directly correlated with
the in vitro results. The most potent compound (23) had an excellent AChE inhibitory
activity having IC50 value of 4.8 nM and TEM analysis confirmed self-induced
inhibition of Aβ1–42 aggregation [56] to the extent of 81.65%. Self-induced Aβ1–42
aggregation is linked to fibril generation that is controlled by formation of β-sheet
pattern which could assist aggregation of Aβ1–42 . The change in the β-sheet pattern
during fibrillo-genesis is mainly stabilized by Glu22/Asp23 and Lys28 residue salt
bridges in Aβ1–42 along with hydrophobic interactions. Compound (23) was found to
be having interactions with amino acids Met35, Leu34, Ile32, Ile31, Gly29, Lys28,
Val24, Asp23, Glu22, Phe20, Ala21, Leu17, Lys16, His13, and other residues
through hydrophobic interactions in the vicinity of Aβ C-terminus, and a strong
78 M. R. Yadav et al.

Fig. 3.10 Binding interactions of compound (23) with AChE. The inhibitor is displayed in ball
and stick model while the key residues are displayed in stick model. Individual colors represent
the residues like Ser200 (cyan), Glu199 (red), and His440 (orange), Tyr70, Tyr121, Tyr334 (blue),
Trp84, Trp279 (hot-pink), Phe330, Phe331 (firebrick), and Asp72, Asp284 (yellow). Stick repre-
sentation is used for other residues with carbons being shown in pink color; (A) depicts interactions
of the active site key residues of AChE with compound (23) and plot (B) is Ligplot representation
(reused with permission source Fig. 5, Umar et al. [56])

hydrogen bond was formed between carbonyl oxygen of carboxylate and N–H of
Lys16 with a distance of 1.8 Å in the docking studies. Moreover, it was observed that
intermolecular H-bonds existed between N–H of amide and oxygen of the hydroxyl
of Glu22 (2.1 Å), N–H of pyrazole ring, and oxygen of hydroxyl of carboxylic acid
group of Asp23 (2.6 Å), and finally another one between N–H of amide and Asp 23
carbonyl carboxylic acid (2.3 Å) (Fig. 3.10). Molecular modeling studies strongly
predict excellent binding ability of compound (23) to the amino acids that interfere
with β-sheets formation, leading to inhibition of Aβ1–42 aggregation, and therefore,
these interactions have a direct correlation with the experimental result of the lead
compound offering activity in the nanomolar range.

3.2.4 Quinoline-Based Scaffolds as Anti-AD Agents

Zaib S. et al. disclosed a series of hybrid derivatives of quinoline-thiosemicarbazones

as inhibitors of cholinesterase for the treatment of AD. Among the reported series,
compound (24) proved to be a promising lead with 0.12 ± 0.02 μM IC50 value in
the in vitro evaluation, which was five times more potent than the standard (i.e.,
galantamine, IC50 = 0.62 ± 0.01 μM) used for the purpose of comparison. Molec-
ular docking studies of the synthesized compound were performed using the PDB
ID: 4BDT [57] structure of hAChE, determined by X-ray crystallography. The most
potent compound (24) showed important interactions like T-shaped π–π interaction
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 79

(6.23 Å) by the quinoline ring with the amino acid residues of the active pocket, and
thiosemicarbazone moiety interacted with Trp86 and formed π-sulfur bond (4.34 Å).
H-bond was observed between Gly122 and the oxygen atom (2.82 Å). Tyr449
(5.86 Å) and Trp439 (4.68 Å) showed π-sulfur interactions with the sulfur atom
of thiosemicarbazone group of compound (24). Oxygen of the methoxyl group inter-
acted with Gly122 (2.82 Å) and formed H-bond, while non-bonding electron pair-π
(2.80 Å) interaction was observed between chloro group and Tyr337. Furthermore,
Glu452 displayed charge–charge attractive interaction (4.93 Å) with the morpholine
nitrogen. It was clearly noted that the deep cleft of the catalytic site was not occupied
by the methoxyl group. The most active compound (24) and Huprine W (Hup W) as
a standard were used for molecular dynamics study. The study involved simulating
the complexes of hAChE (PDB ID: 4BDT) with both the ligands, i.e., HupW and
the selected inhibitor (24), taking the configuration of the docking poses of both
molecules as the initial starting points in an aqueous environment for 30 ns.

H H
H3CO N N
N N
S O
N Cl

(24)

The results (Fig. 3.11) of the simulations were displayed as RMSD and RMSF
values. The standard ligand (HupW) started showing stability from 5 ns onward, with
only a small deviation observed during 2–4.8 ns. The chosen inhibitor (24) showed
stability throughout the simulation period, with only a small deviation during the
period 0.3–0.4 nm. The structure of the complex between the protein and compound
(24) showed only a slight deviation between 14–17 ns and remained stable for the
remaining period of simulation. Figure 3.11 reflects the root mean square fluctuations
of both the inhibitors. A significant pattern of fluctuations was observed for the apo
and holo forms of the protein during the simulation. The apo form of the protein
started showing fluctuation between 0.1–0.3 nm with a rise to 0.4 nm, while the

Fig. 3.11 RMSD and RMSF of the protein (4BDT) amino acid residues in the absence and pres-
ence of HupW and compound (24) during the simulation period of 30 ns (reused with permission
source Fig. 7, Zaib et al. [57])
80 M. R. Yadav et al.

protein-inhibitor complex [hAChE-compound (24)], i.e., the holo form began with a
fluctuation of 0.15 nm, varying between 0.1–0.28 nm with a minor increase. The holo
protein, i.e., hAChE-HupW complex offered 0.10 and 0.30 nm fluctuations during
the entire period of simulation of 30 ns. Both systems demonstrated high stability
and limited level of fluctuations.
Elisabet et al. disclosed some potent BuChE and AChE inhibitors possessing
additional antioxidant activity. Compound (25) from the series emerged as an inter-
esting finding having significantly reduced oxidative stress and neuro-inflammation,
improved basal synaptic efficacy, and much significantly reduced Aβ 42 :Aβ 40 ratio
in the hippocampus. The compound exhibited IC50 values of 1.1 and 7.3 nM against
the enzymes hAChE and hBuChE, respectively [58].

HO O
N NH2
NH N N
H3CO

N Cl

(25)

Molecular modeling studies revealed that the triazole ring showed π-stacking
with the Phe290 side chain while the nitrogen of triazole ring formed H-bond with
Phe288 and Phe295 in hAChE. Additionally, the nitrogen atom of amide group in
compound (25) formed H-bonds with Ser286 as well as residual water present in the
structure, which itself was H-bonded to Phe331. Ether oxygen of capsaicin fragment
of compound (25) and hydroxyl group of Tyr121 makes a water bridge via water 78.
While in case of BuChE, the chloro group of huprine was stacked in acyl-binding
pocket, with Leu286 (O) and Trp231 at 4.0 and 4.1 Å, respectively. The aromatic
system showed π-stacking with Phe329 and Trp231. Moreover, the amino group on
9-position got fitted between His438 and Ser198 (Fig. 3.12).
Novel cyclopentaquinoline derivatives were reported by Czarnecka et al. whereby
they modified the tetrahydroacridine moiety and reported 6-chloronicotinamide
derivative of cyclopentaquinoline (26) possessing EqBuChE and EeAChE inhibitory
activities with IC50 values of 153 and 67 nM, respectively [59]. The compound (26)
was also studied for its molecular interactions with the enzymes using docking studies
on Gold Suite 5.1.

O CH3
N N NH
H
Cl

N
(26)
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 81

Fig. 3.12 Crystal structures of 25-TcAChE and 25-hBuChE complexes (reused with permission
source Fig. 5a, b, Viayna et al. [58])

The docking results clearly suggested that the modified tacrine fragment estab-
lished interactions with the CAS site of the AChE enzyme, in which the cyclopentyl
ring got sandwiched due to π–π stacking with Phe330 and Trp84 and cation-π
interaction. His440 carbonyl carbon made hydrogen bonding with the protonated
nitrogen while Trp279 and Tyr70 residues got into π–π stacking with chloropyri-
dine ring and CH–π interactions with Tyr121 residue, in the AChE PAS site. Docking
of compound (26) with BuChE revealed that the disposition of the cyclopentaquino-
line ring was having similar pattern as was present with AChE, in which Trp82
established π–π stacking and hydrogen bonding with the main chain of His438. The
6-chloronicotinamide fragment of the compound (26) exhibited hydrophobic inter-
actions with Ile356, Pro285, and Tyr282 near the PAS sub-site entrance and hydrogen
bond formation with Tyr332 residue of the main chain.
Karolina et al. reported some novel cyclopentaquinoline-conjugated 9-acridine
carboxylic acid derivatives as cholinesterase inhibitors. Compound (27) offered the
highest inhibitory activity against both the enzymes AChE and BuChE yielding IC50
values of 273.33 and 03.73 nM, respectively [60]. In molecular modeling studies,
it was revealed that the compound (27) possessed dual-binding modes. TRP86 and
TRP286 in parallel disposition showed π-stacking with acridine and modified tacrine
moieties. Oxygen of the amide group got stabilized by the H-bond formation with the
TRY124. Complex of the compound with BuChE showed parallel disposition with
the tacrine fragment, π-stacking with TRP82, and TRP231 and PHE329 residues
were involved in the T-shaped π-stacking with acridine (Fig. 3.13).
82 M. R. Yadav et al.

Fig. 3.13 Compound (27) in the binding pocket of BuChE (reused with permission source Fig. 6,
Czarnecka et al. [59])

N N
H
N
N
H
O

(27)

Safarizadeh et al. formulated a QSAR model for the inhibition of Aβ1–42 peptide
aggregation by the novel 2-arylethenylquinoline derivatives. Molecular modeling
studies were also undertaken to look into the interactions of these novel molecules
with the binding sites of the target [61]. AutoDock 4.2.5.1 was used for performing the
docking studies on Aβ1–42 (PDB 1IYT), and the peptides were analyzed in Discovery
Studio, version 4.5. Results of the docking studies of the designed molecules 2-
arylethenylquinoline with Aβ1–42 peptide showed their binding with amino acid
residues like LEU31, VAL24, MET35, LYS16, ALA21, LEU17, PHE20, and HIS13
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 83

CH3 CH3
N N

N N

N N
N CH3 NO2
CH3

(28) (29)

N N

N N
N CH3 NO2
CH3

(30) (31)

through hydrogen bonds along with hydrophobic interactions. The quinoline ring
N-atom made hydrogen bonding specifically with ALA21 NH group along with
another hydrogen bonding between the methyl group (donor) of diethylamino and
PHE20 NH (acceptor). Moreover, significant level of hydrophobic interactions also
occurred among the electron-donating groups on piperidyl ring with VAL24, ILE31,
and LEU34, as well as of phenyl ring with LEU17, LYS16, and HIS13. Overall,
the investigations revealed that with changes in substituents on the phenyl ring,
substantial changes in the interactions were observed which could directly correlate
with the activity and therefore, while designing novel quinoline compounds for the
inhibition of Aβ1–42 aggregation, the position of electron withdrawing and donating
groups in the aryl ring should be given due importance in order to get more active
agents.
Molecular dynamics was carried out for the most potent compounds (28 and 29)
and the least potent ones (30 and 31) wherein RMSD was used to evaluate the stability
of the Aβ1–42 protein with respect to the defined molecules. The RMSD values were
below 2 Å for Aβ1–42 (RMSD 0.55 Å) with compounds (28 RMSD 0.55 ± 0.09 Å and
29 RMSD0.47 ± 0.04 Å) which got equilibrated after 25 ns while the complexes of
compounds (30 and 31) got stabilized at 32 and 25 ns respectively, which meant that
84 M. R. Yadav et al.

the complexes were less stable and therefore weak in potency with RMSD (1.04 ±
0.03 Å and 1.01 ± 0.04 Å, respectively). The experimental results were in agreement
with the inference obtained from the molecular dynamics studies, as evaluated by
radius of gyration (Rg) plots.
Jiang et al. evaluated some novel PDE2 inhibitors with additional antioxidant
activity [62]. Among the compounds of the designed series, compounds (32 and 33)
exhibited potent PDE2A inhibitory activity with IC50 values of 4.2 and 6.1 nM with
significant level of

N CH3
OH

N N N
R H
N

Cl

(32) R = 4-F; (33) R = 6-F

antioxidant activity [ORAC (Trolox) = 8.4 eq]. Compound (32) offered the least
IC50 value among all of the designed compounds, against PDE2A . In the docking
pose of 32 with PDE2A, the triazoloquinoxaline scaffold showed indirect H-bond
interactions with the Tyr655, Gln859, Gln812, and Tyr827 via three molecules of
water. Direct π-σ and π–π interactions were observed with 6-F group of compound
(33) with Ile826, Phe862, and Phe830 (Fig. 3.14). Hydrophobic interactions were
observed among 2-chlorophenyl group and the neighboring residues Tyr655 and
Leu770. Additionally, 6-F substituted phenolic group was suitably fitted in a pocket
formed by four hydrophobic residues Leu770, Lue774, Phe862, and Ile866.

3.2.5 Coumarin and Chromene-Based Scaffolds as Anti-AD

Agents

Along with cholinesterases, tau protein hyper-phosphorylation, and amyloid-β aggre-

gation, AD is also linked with MAO-B enzyme activation. Changjun Zhang group
reported a series of hybrid hydroxypyridinone-coumarin derivatives with potent
inhibitory activity (IC50 value of 14.7 nM) against the enzyme MAO-B for which
docking studies were also undertaken to discover the enzyme-ligand interactions and
establish SAR for the same [63].
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 85

Fig. 3.14 Binding interactions of PDE2A (PDB ID: 4D09) with compound (33) (reused with
permission source Fig. 2, Jiang et al. [62])

CH3 CH3
HO HO
N N
F
O O O OCH3 O O O O

(34) (35)

CDOCKER program was utilized to undertake molecular docking studies using

the Discovery Studio 2016 software. On the basis of in vitro results, two compounds
(34 and 35) were identified which revealed that the coumarin moiety showed
lipophilic interactions along with hydroxypyridinone toward FAD cofactor and
showed it clearly that hydroxypyridinone moiety acted as a key fragment as a metal
chelator in binding to the MAO-B enzyme substrate. The pyran-carbonyl oxygen
produced two hydrogen bonds with Gln206 and Tyr326 acting as the key compo-
nent for imparting significant inhibitory activity to compound (34), while in case
of compound (35), the benzyloxy fragment occupied the hydrophobic entrance and
interacted with Phe103, Pro104, Leu164, Ile316, and Ile199. The hydroxyl of the
pyridinone interacted by forming hydrogen bond with Gln206, while methyl part of
pyridinone and Tyr435 offered π-alkyl interaction along with FAD which resulted
in providing the highest activity to compound (34).
86 M. R. Yadav et al.

O O O O O O
Br Cl

CH3 NH2
N N
H H O

(36) (37)

Another such series of coumarin derivatives was designed by Rullo et al. in

which 7-benzyloxy-2H-chromen-2-one derivatives (36, 37) were reported possessing
AChE, BuChE, and MAO-B (0.0022 ± 0.0001 μM) inhibitory activities [64]. The
docking was performed on PDB ID: 6O4W using Maestro software package wherein
the heterocyclic core was found to be binding to the AChE in the PAS site, just like
the standard drug donepezil, even though the coumarin ring was not overlapping
the Trp286 residue properly for making π–π interactions. The entire molecule was
stabilized by π-cation and hydrogen bond interactions attaching the basic head to the
entrance of the CAS gorge with Ser293 and Trp286 residues. Docking was repeated
on another PDB ID: 6EY6, and the standard drug galantamine was docked for vali-
dation purpose, but the results were the same. Docking on PDB ID: 6F7Q of hBuChE
showed possible binding of coumarin moiety of compound (36) at PAS Tyr332 and
CAS Trp82 of BuChE, while compound (37) produced a twist which was not suited
to interact with Trp82 resulting in comparatively lower anti-BuChE activity which
was commensurate to the in vitro results.
Palareti et al. have disclosed a novel series of 4-isochromanones composed
of N-benzylpyridinium moiety as potent acetylcholinestrase inhibitors. Compound
(38) among the synthesized compounds has shown potent AChE/BuChE inhibitory
activity with an IC50 value of 0.15 nM [65]. To understand the interactions of the
reported compounds with the active site residues of AChE (PDB ID: 1EVE), molec-
ular docking studies were undertaken. The docking studies revealed that compound
(38) made hydrogen bond interactions between methoxyl groups and Trp286 residue
similar to donepezil, wherein phenylpyridine moiety interacted with the CAS site
and 4-isochromanone moiety interacted with the PAS site of AChE.

CH3

O O
O
O O
H3CO F
O
O N
H3CO .Br

(38) (39)
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 87

Maharajan et al. undertook investigations of 7-propyl-6H-pyranodichromene-

6,8(7H)-dione (39), a novel coumarin inhibitor of Alzheimer’s as a potential drug
acting at triple sites, using various in silico techniques including computational
quantum calculation, molecular dynamics and molecular docking for the inhibition
of β-secretase, glycogen synthase kinase, and acetylcholinesterase enzymes, which
are potential targets for the treatment of AD [66].
Molecular docking studies were performed on the compound (39) taking the
crystal structures of BACE1 (4GID), GSK3β (1Q41), and AChE (4EY6) from
protein data bank. The proteins and ligands structures were optimized and docked
in Schrodinger package-2014 (Maestro, 2014), and the interactions were identified
in PyMol. The most potent compound (39) was docked within the binding sites of
GSK3β, BACE1, and AChE using induced-fit docking offering the docking scores of
−8.51, −8.62, and −10.48, respectively, while the Glide’s free binding energies were
−51.70, −51.83, and −52.71 kcal/M, respectively, which clearly predicted AChE-
biscoumarin complex to be more stable with higher docking score as well as Glide
free energy than the rest of the targets. Moreover, it also revealed that biscoumarin
had strong binding affinity for the CAS site residues Ser203 and His447 of AChE with
interaction distances of 2.4 and 3.2 Å, respectively. Molecular dynamics simulation
was done with leap module and AMBERTOOLS14 with force field AMBER ff14SB
for the newly synthesized coumarin derivatives up to 40 ns, wherein the RMSD
showed similar deviations in the biscoumarin-AChE complex with maximum devi-
ation of ~2.0 Å, making it highly stable, while biscoumarin-BACE1 had a sudden
deviation around 22 and 33 ns due to absence of intermolecular interactions, although
the deviation was ~2.8 Å which could be considered exhibiting high stability. The
radius of gyration (Rg) values for GSK3β, BACE1, and AChE was 21.5, 21, and
~23 Å, respectively, indicating high compactness of the molecule within the proteins.
The results after the MD stimulations of 40 ns showed strong hydrogen bonding
between biscoumarin and the CAS site of AChE with a distance of 1.8 Å, while
in case of GSK3β, some other neighboring interactions were seen, and the active
site interactions disappeared due to the dynamic behavior of the compound. Free
binding energies were determined to be −27.74, −21.19, and −26.43 kcal/M for
AChE, BACE1, and GSK3β, respectively, by selecting 2000 frames from 0 to 40 ns
(Fig. 3.15).
In 2020, Sepehri et al. reported synthesis, characterization, biological evaluation,
and molecular docking of hybrid molecules consisting of coumarin-1,2,3-triazole-
acetamides. They have reported some hybrid compounds containing coumarin scaf-
fold and evaluated them against enzymes like human carbonic anhydrase I & II, α-
glycosidase, α-amylase, acetylcholinesterase, and butyrylcholinesterase, offering Ki
values ranging from 55.38 to 128.63 nM against α-Amy, 590.42–1104.36 nM against
α-Gly, 27.17–1104.36 nM against BuChE, 24.85–132.85 nM against AChE, 508.55–
1284.36 nM against hCA II, and 483.50–1243.04 nM against hCA I. Molecular
modeling was undertaken to investigate the enzyme-ligand interactions. Compound
(40) was found to be the most potent derivative against both the cholinesterase
enzymes. It offered IC50 values of 33.12 and 29.35 nM against BuChE and AChE
enzymes, respectively. The compound interacted with AChE amino acid residues
88 M. R. Yadav et al.

Fig. 3.15 MD simulation trajectories of biscoumarins with AChE, BACE1, and GSK3β for 40 ns.
a RMSD, b radius of gyration, c RMSF of AChE, d RMSF of BACE1, and e RMSF of GSK3β
(reused with permission source Fig. 5, Sivakumar et al. [66])
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 89

Trp84, Phe300, and Phe301 of the CAS region and with Trp279 and Tyr334 of the
PAS sub-site [67].

N F
O O O N O
N
N
H

(40)

The potency of chromones as AChE antagonistic agents has been a key point in
the investigations of anti-Alzheimer’s drug discovery. One such study was carried
out by Prayasee Baruah group wherein they designed new chromone analogs,
mainly cyanochromone (CyC) (41) and aminomethylchromone (AMC) (42). After
performing the docking studies, the compounds were evaluated for their in vitro
activity [68].

O
O N

H2N O CH3
O

(41) CyC (42) AMC

Molecular docking was performed on HSA (PDB ID: 1AO6) and AChE (PDB
ID: 1C2B) proteins taken from Protein Data Bank. The structures were docked using
AutoDock Vina48, while the results were visualized in VMD1.9.352 for visualizing
the molecular interactions of the proteins’ residues with the ligands. With AChE, CyC
was predicted to have a binding score of −35.02 kJ/mol making hydrogen bond
interaction with residues Tyr12 and Tyr337. CyC made hydrophobic interactions
with the amino acid residues Phe338, Gly124, Trp86, and Phe297. In comparison,
AMC made hydrogen bond interactions with residues Tyr124 and Arg296 along with
hydrophobic interactions with Tyr72, Trp286, Ser293, Phe297, and Ile294, offering
a binding score of −33.37 kJ/mol in the modeling studies. Simulation in the PAS
binding was performed in competition with Thioflavin-T that showed that CyC (41)
and AMC (42) both replaced Thioflavin-T non-competitively at the PAS binding
site, and this hypothesis was validated from the results of Achilles Blind Docking
wherein it was seen that π–π stacking of CyC had taken place with the residues Trp
86 and Phe 295, which could be playing important role in the inhibitory activity of
these chromones.
90 M. R. Yadav et al.

3.2.6 Pyrazole-Based Scaffolds as Anti-AD Agents

Taslimi et al. reported synthesis and carbonic anhydrase, α-glycosidase, and

cholinesterase inhibition activities of some pyrazole-phthalazine derivatives [69].
The synthesized compounds were evaluated against α-glycosidase, BuChE, AChE,
hCA I, and hCA II enzymes. To get an insight into the binding interactions of the
synthesized compounds, modeling studies were performed with these enzymes. It
was observed that the dioxo group was critically involved in deciding the potency of
the inhibitory activities of the synthesized compounds. Compound (43) was observed
to be exerting the highest inhibitory action against BuChE and AChE. Docking scores
of compound (43) were found to be −9.91 and −12.31 against AChE and BuChE,
respectively. Further, IC50 values for compound (43) were found to be 94.37 and
98.25 nM against AChE and BuChE enzymes, respectively.

NO2
H
H2N O N
N
O
CF3
N CH3
N
CN O
N

O NH2 H3CO
(43) (44)

In 2021, Zhou et al. reported a series of dihydropyranopyrazole derivatives by

modification of their previously reported compound (R)-LZ77 by using structure-
based designing approach [70]. After suitable modifications, one (44) of the resulting
compounds showed an IC50 value of 41.5 nM for PDE2 inhibitory activity, while
compound (R)-LZ77 exhibited moderate activity against PDE2 with 261.3 nM as
the IC50 value.
Based on the docking studies (Fig. 3.16), it was observed that the π–π stacking
interactions were formed between the phenyl ring of compound (R)-LZ77 and F862,
and H-bonding between cyano group of thepyranopyrazole and Q859. Pyranopy-
razole ring’s oxygen atom established H-bonding with Y655 utilizing the water
molecule as the bridge. A pocket (H-pocket) was formed by the amino acid residues
L809, T805, I870, L770, H773, and F862. The benzyl side chain did not enter the
hydrophobic pocket and remained outside the pocket. That makes it clear that if a
side chain is introduced into the structure of (R)-LZ77 that could occupy this pocket
leading to improvement in the PDE2 inhibitory activity and selectivity in comparison
with the other PDEs.
In the docking studies, it was observed that the oxygen atom of the 5-methoxyl
of the enantiomer (+)-(44) established H-bond with Y827. This bonding improved
the activity of this enantiomer over the derivative having a methyl group at this
position. Position-2 benzyl side chain could enter the H-pocket of the enzyme and
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 91

Fig. 3.16 Binding interactions of (R)-LZ77 with PDE2 as seen in molecular docking studies (upper
part), and the structure-based designing strategy for PDE2 inhibitors (lower structure) (reused with
permission source Fig. 2, Zhou et al. [70])

establish strong hydrophobic interactions with amino acid residues L809, T805,
I870, L770, H773, and F862, improving its inhibitory activity. H-pocket’s size was
large enough to accommodate the full side chain along with its 4-CF3 substituent
(Fig. 3.17a). It is interesting to note that in the case of the other enantiomer (−)-(44),
the pyranopyrazole moiety shifted away from the H-pocket compromising its ability
to make contacts with F862 and Q859 residues. So, the inhibitory activity of (−)-(44)
against PDE2 got drastically reduced.

3.2.7 Benzimidazole and Benzodiazepine Derivatives

as Anti-AD Agents

To study the acetylcholinesterase and butyrylcholinesterase inhibiting potential, Acar

Cevik et al. designed compounds incorporating benzimidazole and triazole rings in
the same chemical moiety and performed molecular modeling studies to find out the
binding mechanism of these compounds with these proteins. Compounds (45 and
46) which contained 3,4-dihydroxyphenyl along with 5(6)-chloro substituent in the
benzimidazole ring and alkyl substituent on the triazole ring were discovered to be
92 M. R. Yadav et al.

Fig. 3.17 Binding modes of

both the enantiomers of
compound (44) in the
docking studies with PDE2.
(A) (+)-isomer, (B)
(−)-isomer (source Fig. 4,
Zhou et al. [70])

effective AChE inhibitors. The Ki and IC50 values of compounds (45 & 46) were
found out to be 26.2 nM and 31.9 ± 0.1 nM, respectively [71].

OH
H
Cl N N N OH

N N S
R O
(45) R = CH3; (46) R = C2H5

Docking studies of compounds (45 & 46) were undertaken using the 3D-structure
(PDB ID: 4EY7) for AChE. It was found that the benzimidazole ring interacted with
Trp286 of PAS through lipophilic interaction, while the polar dihydroxyphenyl-
triazolylthioethanone side chain was bound to Trp86 residue of the CAS sub-site.
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 93

The results of 45 and 46 bound to the enzyme produced six mutual interactions,
wherein 5(6)-chlorobenzimidazole ring and 3,4-dihdroxyphenyl ring offered π–π
interactions with Trp286 and Trp86, respectively, and triazole N-atom interacted
with Trp24 via H-hydrogen bonding. 3,4-dihdroxyphenyl ring was also found to be
interacting with Gly120 and Tyr133 through polar interaction by acting as H-bond
donor, and Tyr133 acting as acceptor. 3-Hydroxyl of phenyl ring also made hydrogen
bonding with Glu202 residue through its carbonyl group.
An interesting series of benzodiazepine-1,2,3-triazole derivatives was reported
possessing cholinesterase inhibitory activity in which the compounds inhib-
ited butyrylcholinesterase (BuChE) enzyme but not acetylcholinesterase. Two
compounds (47 and 48) were identified as sub-micromolar BuChE inhibitors offering
IC50 values of 0.2 and 0.4 μM, respectively [72]. The docking studies were performed
using AChE-donepezil complex (PDB ID: 4EY7) and BuChE-tacrine complex (PDB
ID: 4BDS) on Autodock (ver.4.2.6). The docking results of compounds (47 and
48) vis-à-vis donepezil and tacrine indicated strong interactions with the ligands
donepezil and tacrine. The standard ligands were re-docked into the active sites of
the enzymes for the purpose of validation of the process. RMSD of the re-docked
ligands in comparison with the co-crystallized structures was 1.12 and 0.062 Å,
respectively, for tacrine and donepezil. The results obtained for compounds (47 and
48) clearly suggested that these structures could be ideal inhibitors of BuChE as
they showed multiple interactions within its binding site (PAS). The dimethyl group
of compound (47) produced π-alkyl interactions with Trp286 an aromatic residue
present in PAS. Another π-anion interaction was seen between compound (47) phenyl
ring and Asp74 residue, at the CAS site of AChE. π–π stacking interaction was
seen between 3-fluorobenzyl group and Trp86, along with H-bond with Trp133 and
Ala127. Additional hydrogen bonds were observed between the nitrogen of the tria-
zole ring with both His447 and Ser203 residues. Docking studies performed on
compound (48) indicated that the molecule was comfortably placed inside the active
site of the enzyme with free binding energy of −8.6 kcal/M. The ether linkage oxygen
atom formed H-bonds with Ser198 and His438 amino acid residues, while the phenyl
group linker was involved with Glu197 in a π-anion interaction and with Trp82 in π–
π interaction. The 4-methyl-subtituted benzyl ring produced another π–π stacking
interaction with Tyr332 and Phe329.

H3C H3C
CH3 H CH3
H N
N
CH3
N N N N
O N O
N N H N
H F O
O

(47) (48)
94 M. R. Yadav et al.

3.2.8 Thiazole Containing Compounds as Anti-AD Agents

Thiazole-piperazine hybrids as a novel class of compounds were developed by

Osmaniye et al. to combat Alzheimer’s disease. On cholinesterase enzymes, the
inhibitory potential of compound (49) was found to be significant with IC50 value of
0.0317 μM against AChE, which was supported by the in silico studies [73].

H3CO

S
N
HN N
N N F

(49)

Molecular docking studies were performed on compound (49) and the standard
drug donepezil using the crystal structure having PDB ID: 4EY7, wherein it was
revealed that compound (49) was showing binding interactions to CAS and PAS
sub-sites of AChE in a disposition similar to donepezil to Trp86 and Trp286, respec-
tively, while the standard donepezil formed interactions with Trp286 and Phe295
of the PAS along with Trp86 of the CAS. Specifically, the thiazole ring of (49)
formed π–π interaction with Trp286 indole ring system while the indole ring of
Trp86 and 4-fluorophenyl ring offered π–π interactions. The piperidine phenyl ring
and Tyr341 phenyl ring exhibited another π–π interaction. The nitrogen atoms of
imine and piperidine ring jointly formed the basic centers wherein the imine group
formed hydrogen bond with Tyr341, and the piperidine ring formed H-bond with
hydroxyl group of Tyr337. Lastly, oxygen of 4-methoxyphenyl and carbonyl of
Leu289 established hydrogen bond interactions resulting in manifestation of potent
AChE inhibitory activity to the compound (49).

3.2.9 Alkylamine Linked Derivatives as Anti-AD Agents

Synthesis and biological activity of phthalide tertiary amine derivatives have been
reported by Luo L et al. as AChE inhibitors. The reported compounds demonstrated
high selectivity and potency in AD [74]. Compound (50) offered the highest potency
with an IC50 value of 2.66 nM against AChE. The compound (50) was bound to PAS
as well as to CAS sub-sites as revealed in the docking studies. It was found to have
excellent in vitro BBB permeation. Moreover, in the animal studies, the compound
could exhibit reversal of scopolamine-induced memory deficit.
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 95

O
H3CO
O
H3CO CH3
N
3

(50)

Some novel ferulic acid derivatives have been reported by Lan et al. for the
treatment of Alzheimer’s disease [75]. Most of the synthesized compounds exhib-
ited potent AChE enzyme inhibition and self-induced β-amyloid (Aβ) aggregation
activity with good antioxidative potential. Compound (51) offered IC50 values of
0.66 μM for BuChE and 9.7 nM for AChE, the lowest ones among all the compounds
of the series. At a concentration of 20 nM, it offered 49.2% inhibition of Aβ aggre-
gation. This was equivalent to 1.26 trolox antioxidant activity. Further, modeling
and kinetic studies revealed that compound (51) interacted with CAS and PAS sub-
sites of AChE simultaneously offering a mixed type of inhibition. It was found to
show good BBB permeation potential in the PAMBA-BBB assay. All of the above
observations proved that compound (51) could be a potential multifunctional lead
for inhibiting ChE enzymes.

O
H3CO
O N O
2
CH3 CH3
HO O N
CH3

(51)

A combined study of molecular docking and QSAR and molecular dynamics

of some AChE and BChE inhibitors have been reported by Daoud et al.
[15]. The study was initially carried out on 36 compounds belonging to 4-
[(diethylamino)methyl]phenyl ether class of derivatives. Compounds (52 and 53)
have been identified as novel anti-Alzheimer’s agents for further development on
the basis of this study. Compound (52) offered an IC50 value of 0.084 μM against
the enzyme AChE interacting with it via TYR34, TRP84, and TRP279 residues, and
compound (53) with IC50 value 0.0091 μM was found to interact with BuChE.
96 M. R. Yadav et al.

H3C H3 C
N N
CH3 CH3 CH3
O N O N
3 5
N
CH3
CH3

(52) (53)

Begum et al. carried out in vitro as well as in silico studies of N-phthaloylglycine

amide derivatives as BuChE inhibitors [76]. Schotten Baumann reaction was used to
synthesize these amide derivatives. The compounds displayed potent BuChE inhi-
bition activity in the in vitro evaluation. When compared with galantamine (IC50
value of 6.6 μM), compound (54) (IC50 value of 6.5 μM) was found to be equipo-
tent to galantamine and the most active compound of the series. Molecular docking
studies were undertaken for these compounds to unravel their high order of biological
activity. A majority of the compounds of the series were fitting well in the BuChE
enzyme binding sub-sites. The heterocyclic moiety of compound (54) established
H-bonding interactions with TYR332 residue of the binding site. Further, it was
found that π–π interactions were stabilizing the compound in the enzyme cavity
(Fig. 3.18).

Fig. 3.18 Compound (54) docked within the binding gorge of BuChE (reused with permission
source Fig. 3, Begum et al. [76])
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 97

CH3
O O
N
N
CH3
O

(54)

Xie et al. reported a series of hybrid molecules on the basis of clorgyline-rasagiline

structure and assessed them for inhibition of amyloid-β aggregation and monoamine
oxidase enzyme [77]. Among the synthesized compounds, a majority of them showed
good hMAO-B inhibitory activity, but the chroman derivative (55) showed excep-
tionally good activity as hMAO-B inhibitor (IC50 value of 4 nM) and Aβ aggregation
inhibitor (40.78% inhibition at 25 μM concentration).

2 CH
O O 5N

CH3

(55)

Docking studies were performed to understand the binding interactions occur-

ring between compound (55) and hMAO-B (2V61) active site. It was observed
that the ligand showed arene-H interactions with residue Tyr435 and van der Waals
and hydrophobic interactions with leu717, Cys172, Ile198, Ile199, Ile316, Tyr326,
Trp119, and Phe168 amino acid residues.
Gobec et al. have reported a series of tryptophan derivatives as selective BuChE
inhibitors. Among the synthesized compounds of the series, compound (56) offered
IC50 value of 2.8 nM against BuChE enzyme, the lowest among the series [78]. Molec-
ular modeling studies were performed in the active site of the BuChE enzyme in order
to assess the binding interactions of the compound. Modeling studies revealed that
the compound was engaged into a l–l interaction with residue Trp231 at a distance
of 4.3 Å at the acyl-binding pocket of the enzyme BuChE and also established
H-bonding with Pro285 amino acid residue. Amino

NH

H N
N
H3C N
H
O

(56)
98 M. R. Yadav et al.

Fig. 3.19 Binding of compound (56) in the active binding site of BuChE (reused with permission
source Fig. 6, Meden et al. [78])

acid residues Phe329, Gly116, and Gly117 were bound through van der Waals
interactions with the compound (56), and residues Gln119, Ala277, Thr284, Ser287,
and Asn289 made contacts through van der Waals interactions (Fig. 3.19).

3.2.9.1 Miscellaneous Compounds as Anti-AD Agents

Amide-based drugs have the potential to interact with various enzymes and recep-
tors and generate a biological response. Hence, amide and carbamate scaffolds have
become a popular topic in medicinal chemistry. Jiang and colleagues investigated
the use of cannabidiol (CBD) and carbamate hybrids as selective inhibitors of ChE
enzymes. They found that CBD could fit well into the groove of BuChE through
several π–π interactions and hydrogen bonds. Earlier studies had shown that a
hydroxyl group of CBD occupied enough space for interaction with Thr120. Docking
studies indicated that the carbamate carbonyl group formed hydrogen bonds with the
residues in the enzymes, and the nitrogen atom in the molecule was having crucial
interactions with the binding site of ChEs. Based on these findings, the researchers
designed 17 new compounds using a structural reassembly approach. Compounds
(57) demonstrated high potency (IC50 = 5.3 nM) and selectivity for the inhibi-
tion of BuChE. The compound also showed good properties such as BBB pene-
tration, pseudo-irreversible inhibition, antioxidant activity, safety, and neuroprotec-
tion. Another compound, (58), also demonstrated nanomolar inhibition of eqBuChE
(IC50 = 7.3 nM) and showed better hBuChE inhibition compared to rivastigmine.
The molecular docking studies showed that phenyl halogen atom improved the lipid
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 99

solubility and membrane permeability, increasing the absorption and transport of the
drug. The substitution of chlorine or bromine atoms increased hydrophobic interac-
tions and π–π interactions with Trp430 and Met437, enhancing the BuChE inhibitory
activity further [79].

CH3 O
H H
F N N F
O N
CH3 NH
R F N N F
H2C CH3 CH3
HO
(57) R = Cl; (58) R = Br (59)

Sayyad et al. reported some biaryl guanidine derivatives and their biological eval-
uation for the treatment of Alzheimer’s by inhibition of β-secretase enzyme [80]. The
authors conducted virtual screening to predict compounds active against β-secretase.
On the basis of virtual screening results, the authors synthesized 13 compounds and
evaluated them for their biological activities through in vitro and in vivo experi-
ments. Compound (59) was found to be the most active compound (IC50 value of
97 ± 0.91 nM) in the series. FRET assay for arresting β-secretase enzyme activity
(99%) also indicated the activity potential of the compound. The compound contains
fluorine atoms in its structure, which were responsible for improving its bioavail-
ability in the enzyme’s active domain of the flap region. In the Morris water maze
novel object recognition test, the compound (59) was found to improve the scores
significantly (p < 0.05).
Das et al. have reported dihyroactinidiolide (DA) (60) as a novel anti-Alzheimer’s
agent. Compound (60) has been synthesized from β-ionone using oxidation method.
From the in vitro studies, it has been revealed that this compound showed potent
AChE inhibition with IC50 value of 34.03 nM [81]. Other than this, compound (60)
was reported to possess DPPH and nitric oxide scavenging and metal chelating
activities with IC50 values of 50 nM and about 270 nM, respectively. The compound
showed no cytotoxicity toward N2a cells.

H3C CH3

O
O
CH3

(60)

Molecular modeling studies also supported the analysis and showed that the
compound (60) was involved in four hydrogen bond interactions through two O atoms
of the lactone ring with residues GLY118, GLY119, and SER200 in the active site. To
100 M. R. Yadav et al.

decode the DA structure–activity relationship, a standard compound γ-butyrolactone

reported to increase acetylcholine levels in mice and rat, and γ-hydroxybutyrate
levels in human brain were docked. From the docking studies, it was revealed that
DA was also involved in four hydrogen bond interactions through its O and H groups
with residues PHE288 and ARG289 in the AChE active site. At a concentration of
270 nM, DA was also reported to have de-aggregation and anti-aggregation poten-
tial on Aβ25–34 fragment. In the docking studies, it offered a free binding energy
of 3.62 kcal/mol, and it established hydrogen bonding with LYS28 residue of the
enzyme.
Kumar et al. studied naphthofuran derivatives as potential inhibitors against GSK-
3β and BACE1 through MM-PBSA binding energy analysis, molecular dynamics,
and molecular docking studies. In comparison with the previously reported inhibitors,
the two newly reported compounds NS7 (61) and NS9 (62) and exhibited much better
binding affinities. Further, MM-PBSA analysis and molecular dynamics demon-
strated that the binding interaction between these compounds and the enzymes was
mainly due to hydrophobic interactions. This study also revealed that these naphtho-
furan derivatives have the potential to act as dual GSK-3β and BACE1 inhibitors by
causing the enzymes to adopt a closed conformation and become inactive [82].

OCH3 H3CO OCH3

H3CO
OCH3 OCH3
NO2
O

O
O O
O O

(61) NS7 (62) NS9

Hu et al. reported dual AChE and PARP-1 inhibitors for the management of AD
[83]. A group of 863 PARP-1 inhibitors possessing IC50 ≤ 10 μM was selected by
the authors in order to identify compounds with high affinity for AChE using virtual
screening technique. Five crystal structures of AChE having some known inhibitors
were utilized for virtual screening by applying same parameters on them. Using
molecular dynamics approach, the complexes were also studied for their dynamic
stability. Among the findings, compounds CID57390505 (63) and CID71605390
(64) attracted the spotlight by exhibiting high stability in the complex form with
the enzyme AChE and high affinity for it in the dynamics and docking studies,
respectively.
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 101

O
O
HN
NH
HN
O
N
O
O
HN CH3

O
O
N

CID57390505 CID1605390

(63) (64)

Hassan et al. used molecular docking and molecular dynamics to study the binding
of Solanezumab and AZD3293 (65) to certain proteins like amyloid-β (Aβ) and β-
secretase (BACE1) involved in the pathogenicity of AD [84]. The results indicated
that hydrogen bonding was involved in the binding of AZD3293 (65) to BACE in
its active site through the amino acid residues Lys107 and Asp32 having distances
of 2.68 and 2.95 Å, respectively. Whereas, solanezumab interacted with Lys16 and
Asp23 of Aβ peptide and formed hydrogen bond with heavy chains of Ser33 and
Asp96, respectively. In case of binding of BACE to solanezumab, Asp96 formed H-
bond with Lys16 with a distance of 2.82 Å, while Asp23 formed two hydrogen bonds
with Ser33 having distances of 2.78 and 3.00 Å. AZD3293 offered better results
in molecular dynamics offering RMSD fluctuation of 0.2 nm in comparison with
0.7 nm offered by solanezumab. All these results indicated that AZD3293 possessed
an advantage over solanezumab for the management of Alzheimer’s disease.

H2N
CH3 CH3
N
N
H3C

N
AZD3293

(65)
102 M. R. Yadav et al.

3.3 Conclusion

Among the main causes of deaths due to dementia, Alzheimer’s is the most common
one throughout the world. In the current scenario, around 50 million people world-
wide suffer from dementia, and as per AD (Alzheimer’s Dement) report by 2050,
the number is bound to jump to more than 150 million. Many targets have been
identified for the treatment of AD, but the exact causative factors responsible for the
disease are not yet fully known, though sustained efforts are being made to unravel
the mystery behind the disease. Computational techniques have proved of immense
help to the researchers for discovering new drugs. These techniques are providing
information about molecular interactions, ligand-enzyme interactions, stability of
the molecular complexes formed after ligand-receptor interactions, and physico-
chemical properties of drugs. The present chapter provides a comprehensive infor-
mation about the reports on the use of computational studies using MTDLs as hybrid
molecules with a potential to be developed as future anti-AD drugs. Tacrine was the
first approved drug which has been withdrawn from the market, but as discussed in
this chapter, tacrine structure has been widely exploited for the discovery of novel
AChE and BuChE inhibitors using molecular hybridization technique. It has been
seen that heterocycles, fused or isolated, such as indole, quinoline, pyrimidine, and
coumarin yielded potent AChE and BuChE inhibitors to be developed further as
potential anti-AD drugs. Most of the hybrid molecules were binding to the active
sites of AChE and BuChE enzymes and thus acted as inhibitors as indicated by these
molecular modeling studies. Quinoline containing moieties were found to be good
inhibitors of Aβ peptide which also exhibited good antioxidant properties. Some
nitrogen containing heterocycles like pyrazole and pyrimidine showed additionally
potent phosphodiesterase inhibitory activity.
In the changing paradigm, molecular modeling techniques are proving invaluable
tools in the drug discovery process. In summary, this chapter has discussed some
important findings involving molecular modeling techniques reported in literature
for the discovery of MTDLs as anti-AD agents. The reported leads could pave the
way for the discovery of new molecular entities useful for the treatment/cure of
Alzheimer’s disease.

References

1. Alzheimer’s Disease Facts and Figures (2020) Alzheimer’s Dement 16:391–460

2. Alzheimer’s Disease Facts and Figures (2021) Alzheimer’s Dement 17:327–406
3. Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, Fagan AM, Iwatsubo T, Jack
CR Jr, Kaye J, Montine TJ, Park DC (2011) Toward defining the preclinical stages of
Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s
Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement
7(3):280–292
4. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR Jr, Kawas CH, Klunk WE,
Koroshetz WJ, Manly JJ, Mayeux R, Mohs RC (2011) The diagnosis of dementia due to
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 103

Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s

Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement
7(3):263–269
5. Briggs R, Kennelly SP, O’Neill D (2016) Drug treatments in Alzheimer’s disease. Clin Med
16(3):247
6. Reiss AB, Arain HA, Stecker MM, Siegart NM, Kasselman LJ (2018) Amyloid toxicity in
Alzheimer’s disease. Rev Neurosci 29(6):613–627
7. Forloni G, Balducci C (2018) Alzheimer’s disease, oligomers, and inflammation. J Alzheimers
Dis 62(3):1261–1276
8. Naseri NN, Wang H, Guo J, Sharma M, Luo W (2019) The complexity of tau in Alzheimer’s
disease. Neurosci Lett 13(705):183–194
9. Kabir MT, Uddin MS, Mamun AA, Jeandet P, Aleya L, Mansouri RA, Ashraf GM, Mathew
B, Bin-Jumah MN, Abdel-Daim MM (2020) Combination drug therapy for the management
of Alzheimer’s disease. Int J Mol Sci 21(9):3272
10. Yang P, Sun F (2021) Aducanumab: the first targeted Alzheimer’s therapy. Drug Discover
Therapeut 15(3):166–168
11. Jeffrey C, Garam L, Travis M, Aaron R, Kate Z (2017) Alzheimer’s disease drug development
pipeline: 2017. Alzheimer’s & Dement: Transl Res Clin Interv 3(3):367–384
12. Athar T, Al Balushi K, Khan SA (2021) Recent advances on drug development and emerging
therapeutic agents for Alzheimer’s disease. Mol Biol Rep 48(7):5629–5645
13. Oset-Gasque MJ, Marco-Contelles J (2018) Alzheimer’s disease, the “one-molecule, one-
target” paradigm, and the multitarget directed ligand approach. ACS Chem Neurosci 9(3):401–
403
14. Murumkar PR, Le L, Truong TN, Yadav MR (2011) Determination of structural requirements of
influenza neuraminidase type A inhibitors and binding interaction analysis with the active site
of A/H1N1 by 3D-QSAR CoMFA and CoMSIA modeling. MedChemComm. 2(8):710–719
15. Daoud I, Melkemi N, Salah T, Ghalem S (2018) Combined QSAR, molecular docking and
molecular dynamics study on new Acetylcholinesterase and Butyrylcholinesterase inhibitors.
Comput Biol Chem 1(74):304–326
16. Murumkar PR, Sharma MK, Gupta P, Patel NM, Yadav MR (2022) Selection of suitable protein
structure from Protein Data Bank: an important step in structure based drug design studies.
Mini Rev Med Chem
17. Murumkar PR, Giridhar R, Yadav MR (2008) 3D-quantitative structure-activity relationship
studies on benzothiadiazepine hydroxamates as inhibitors of tumor necrosis factor-α converting
enzyme. Chem Biol Drug Des 71(4):363–373
18. DasGupta S, Murumkar PR, Giridhar R, Yadav MR (2009) Studies on novel 2-imidazolidinones
and tetrahydropyrimidin-2 (1H)-ones as potential TACE inhibitors: design, synthesis, molec-
ular modeling, and preliminary biological evaluation. Bioorg Med Chem 17(10):3604–3617
19. Sharma MK, Machhi J, Murumkar P, Yadav MR (2018) New role of phenothiazine derivatives
as peripherally acting CB1 receptor antagonizing anti-obesity agents. Sci Rep 8(1):1–8
20. Murumkar PR, Gupta SD, Zambre VP, Giridhar R, Yadav MR (2009) Development of predictive
3D-QSAR CoMFA and CoMSIA models for β-aminohydroxamic acid-derived tumor necrosis
factor-α converting enzyme inhibitors. Chem Biol Drug Des 73(1):97–107
21. Murumkar PR, Zambre VP, Yadav MR (2010) Development of predictive pharmacophore
model for in silico screening, and 3D QSAR CoMFA and CoMSIA studies for lead optimization,
for designing of potent tumor necrosis factor alpha converting enzyme inhibitors. J Comput
Aided Mol Des 24:143–156
22. Sengupta P, Puri CS, Chokshi HA, Sheth CK, Midha AS, Chitturi TR, Thennati R, Murumkar
PR, Yadav MR (2011) Synthesis, preliminary biological evaluation and molecular modeling
of some new heterocyclic inhibitors of TACE. Eur J Med Chem 46(11):5549–5555
23. Murumkar PR, Sharma MK, Shinde AC, Bothara KG (2013) Three-dimensional quantita-
tive structure–activity relationship CoMFA/CoMSIA on pyrrolidine-based tartrate diamides as
TACE inhibitors. Med Chem Res 22:4192–4201
104 M. R. Yadav et al.

24. Murumkar PR, Sharma MK, Giridhar R, Yadav MR (2015) Virtual screening-based identifi-
cation of lead molecules as selective TACE inhibitors. Med Chem Res 24:226–244
25. Sarkate AP, Murumkar PR, Lokwani DK, Kandhare AD, Bodhankar SL, Shinde DB, Bothara
KG (2015) Design of selective TACE inhibitors using molecular docking studies: synthesis
and preliminary evaluation of anti-inflammatory and TACE inhibitory activity. SAR QSAR
Environ Res 26(11):905–923
26. Murumkar PR, Ghuge RB, Chauhan M, Barot RR, Sorathiya S, Choudhary KM, Joshi KD,
Yadav MR (2020) Recent developments and strategies for the discovery of TACE inhibitors.
Expert Opin Drug Discov 15(7):779–801
27. Sharma MK, Murumkar PR, Giridhar R, Yadav MR (2015) Exploring structural requirements
for peripherally acting 1, 5-diaryl pyrazole-containing cannabinoid 1 receptor antagonists for
the treatment of obesity. Mol Divers 19:871–893
28. Roy K (ed) (2018) Computational modeling of drugs against Alzheimer’s disease. Springer,
New York
29. Barmade M, Murumkar P, Kumar Sharma M, Shingala K, Giridhar R, Ram Yadav M (2015)
Discovery of anti-malarial agents through application of in silico studies. Comb Chem High
Throughput Screen 18(2):151–87
30. Yadav MR, Barmade MA, Chikhale RV, Murumkar PR (2018) Computational modelling of
kinase inhibitors as anti-Alzheimer agents. Comput Model Drugs Against Alzheimer’s Disease
347–417
31. Patel DV, Patel NR, Kanhed AM, Patel SP, Sinha A, Kansara DD, Mecwan AR, Patel SB,
Upadhyay PN, Patel KB, Shah DB (2019) Novel multitarget directed triazinoindole derivatives
as anti-Alzheimer agents. ACS Chem Neurosci 10(8):3635–3661
32. Khambete M, Murumkar P, Kumar A, Darreh-Shori T, De S, Yadav MR, Degani MS. Article
details Pyrazoline containing molecules as multifunctional agents in Alzheimer’s disease
33. Shidore M, Machhi J, Shingala K, Murumkar P, Sharma MK, Agrawal N, Tripathi A,
Parikh Z, Pillai P, Yadav MR (2016) Benzylpiperidine-linked diarylthiazoles as potential
anti-Alzheimer’s agents: synthesis and biological evaluation. J Med Chem 59(12):5823–5846
34. Kanhed AM, Patel DV, Patel NR, Sinha A, Thakor PS, Patel KB, Prajapati NK, Patel KV, Yadav
MR (2022) Indoloquinoxaline derivatives as promising multi-functional anti-Alzheimer agents.
J Biomol Struct Dyn 40(6):2498–2515
35. Patel KB, Patel DV, Patel NR, Kanhed AM, Teli DM, Gandhi B, Shah BS, Chaudhary BN,
Prajapati NK, Patel KV, Yadav MR (2022) Carbazole-based semicarbazones and hydrazones
as multifunctional anti-Alzheimer agents. J Biomol Struct Dyn 40(20):10278–10299
36. Patel DV, Patel NR, Kanhed AM, Teli DM, Patel KB, Gandhi PM, Patel SP, Chaudhary BN,
Shah DB, Prajapati NK, Patel KV (2020) Further studies on triazinoindoles as potential novel
multitarget-directed anti-Alzheimer’s agents. ACS Chem Neurosci 11(21):3557–3574
37. Machhi J, Sinha A, Patel P, Kanhed AM, Upadhyay P, Tripathi A, Parikh ZS, Chruvattil R, Pillai
PP, Gupta S, Patel K (2016) Neuroprotective potential of novel multi-targeted isoalloxazine
derivatives in rodent models of Alzheimer’s disease through activation of canonical Wnt/β-
catenin signalling pathway. Neurotox Res 29:495–513
38. Przybyłowska M, Dzierzbicka K, Kowalski S, Demkowicz S, Daśko M, Inkielewicz-Stepniak
I (2022) Design, synthesis and biological evaluation of novel N-phosphorylated and O-
phosphorylated tacrine derivatives as potential drugs against Alzheimer’s disease. J Enzyme
Inhib Med Chem 37(1):1012–1022
39. Yao H, Uras G, Zhang P, Xu S, Yin Y, Liu J, Qin S, Li X, Allen S, Bai R, Gong Q (2021)
Discovery of novel tacrine-pyrimidone hybrids as potent dual AChE/GSK-3 inhibitors for the
treatment of Alzheimer’s disease. J Med Chem 64(11):7483–7506
40. Ozten O, Kurt BZ, Sonmez F, Dogan B, Durdagi S (2021) Synthesis, molecular docking and
molecular dynamics studies of novel tacrine-carbamate derivatives as potent cholinesterase
inhibitors. Bioorg Chem 1(115):105225
41. Chufarova N, Czarnecka K, Skibiński R, Cuchra M, Majsterek I, Szymański P (2018)
New tacrine–acridine hybrids as promising multifunctional drugs for potential treatment of
Alzheimer’s disease. Arch Pharm 351(7):1800050
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 105

42. Li G, Hong G, Li X, Zhang Y, Xu Z, Mao L, Feng X, Liu T (2018) Synthesis and activity
towards Alzheimer’s disease in vitro: tacrine, phenolic acid and ligustrazine hybrids. Eur J
Med Chem 25(148):238–254
43. Derabli C, Boulebd H, Abdelwahab AB, Boucheraine C, Zerrouki S, Bensouici C, Kirsch G,
Boulcina R, Debache A (2020) Synthesis, biological evaluation and molecular docking studies
of novel 2-alkylthiopyrimidino-tacrines as anticholinesterase agents and their DFT calculations.
J Mol Struct 5(1209):127902
44. Zhu J, Yang H, Chen Y, Lin H, Li Q, Mo J, Bian Y, Pei Y, Sun H (2018) Synthesis, pharma-
cology and molecular docking on multifunctional tacrine-ferulic acid hybrids as cholinesterase
inhibitors against Alzheimer’s disease. J Enzyme Inhib Med Chem 33(1):496–506
45. Wieckowska A, Wichur T, Godyń J, Bucki A, Marcinkowska M, Siwek A, Wieckowski K,
Zareba P, Knez D, Głuch-Lutwin M, Kazek G (2018) Novel multitarget-directed ligands aiming
at symptoms and causes of Alzheimer’s disease. ACS Chem Neurosci 9(5):1195–1214
46. Makhaeva GF, Kovaleva NV, Boltneva NP, Lushchekina SV, Rudakova EV, Stupina TS, Teren-
tiev AA, Serkov IV, Proshin AN, Radchenko EV, Palyulin VA (2020) Conjugates of tacrine and
1, 2, 4-thiadiazole derivatives as new potential multifunctional agents for Alzheimer’s disease
treatment: Synthesis, quantum-chemical characterization, molecular docking, and biological
evaluation. Bioorg Chem 1(94):103387
47. Shaikh S, Pavale G, Dhavan P, Singh P, Uparkar J, Vaidya SP, Jadhav BL, Ramana MM (2021)
Design, synthesis and evaluation of dihydropyranoindole derivatives as potential cholinesterase
inhibitors against Alzheimer’s disease. Bioorg Chem 1(110):104770
48. He F, Chou CJ, Scheiner M, Poeta E, Yuan Chen N, Gunesch S, Hoffmann M, Sotriffer C,
Monti B, Maurice T, Decker M (2021) Melatonin-and ferulic acid-based HDAC6 selective
inhibitors exhibit pronounced immunomodulatory effects in vitro and neuroprotective effects
in a pharmacological Alzheimer’s disease mouse model. J Med Chem 64(7):3794–3812
49. Ghamari N, Dastmalchi S, Zarei O, Arias-Montaño JA, Reiner D, Ustun-Alkan F, Stark H,
Hamzeh-Mivehroud M (2020) In silico and in vitro studies of two non-imidazole multiple
targeting agents at histamine H3 receptors and cholinesterase enzymes. Chem Biol Drug Des
95(2):279–290
50. Lee HY, Fan SJ, Huang FI, Chao HY, Hsu KC, Lin TE, Yeh TK, Lai MJ, Li YH, Huang HL,
Yang CR (2018) 5-Aroylindoles act as selective histone deacetylase 6 inhibitors ameliorating
Alzheimer’s disease phenotypes. J Med Chem 61(16):7087–7102
51. Lozinskaya NA, Babkov DA, Zaryanova EV, Bezsonova EN, Efremov AM, Tsymlyakov MD,
Anikina LV, Zakharyascheva OY, Borisov AV, Perfilova VN, Tyurenkov IN (2019) Synthesis
and biological evaluation of 3-substituted 2-oxindole derivatives as new glycogen synthase
kinase 3β inhibitors. Bioorg Med Chem 27(9):1804–1817
52. Manzoor S, Prajapati SK, Majumdar S, Raza MK, Gabr MT, Kumar S, Pal K, Rashid H, Kumar
S, Krishnamurthy S, Hoda N (2021) Discovery of new phenyl sulfonyl-pyrimidine carboxylate
derivatives as the potential multi-target drugs with effective anti-Alzheimer’s action: design,
synthesis, crystal structure and in-vitro biological evaluation. Eur J Med Chem 5(215):113224
53. Zhang C, Zhou Q, Wu XN, Huang YD, Zhou J, Lai Z, Wu Y, Luo HB (2018) Discovery of novel
PDE9A inhibitors with antioxidant activities for treatment of Alzheimer’s disease. J Enzyme
Inhib Med Chem 33(1):260–270
54. Ghobadian R, Nadri H, Moradi A, Bukhari SN, Mahdavi M, Asadi M, Akbarzadeh T,
Khaleghzadeh-Ahangar H, Sharifzadeh M, Amini M (2018) Design, synthesis, and biological
evaluation of selective and potent Carbazole-based butyrylcholinesterase inhibitors. Bioorg
Med Chem 26(17):4952–4962
55. Kumar J, Gill A, Shaikh M, Singh A, Shandilya A, Jameel E, Sharma N, Mrinal N, Hoda N,
Jayaram B (2018) Pyrimidine-triazolopyrimidine and pyrimidine-pyridine hybrids as potential
acetylcholinesterase inhibitors for Alzheimer’s disease. ChemistrySelect 3(2):736–747
56. Umar T, Shalini S, Raza MK, Gusain S, Kumar J, Seth P, Tiwari M, Hoda N (2019) A multifunc-
tional therapeutic approach: synthesis, biological evaluation, crystal structure and molecular
docking of diversified 1H-pyrazolo [3, 4-b] pyridine derivatives against Alzheimer’s disease.
Eur J Med Chem 1(175):2–19
106 M. R. Yadav et al.

57. Zaib S, Munir R, Younas MT, Kausar N, Ibrar A, Aqsa S, Shahid N, Asif TT, Alsaab HO, Khan
I (2021) Hybrid quinoline-thiosemicarbazone therapeutics as a new treatment opportunity for
Alzheimer’s disease-synthesis, in vitro cholinesterase inhibitory potential and computational
modeling analysis. Molecules 26(21):6573
58. Viayna E, Coquelle N, Cieslikiewicz-Bouet M, Cisternas P, Oliva CA, Sánchez-López E,
Ettcheto M, Bartolini M, De Simone A, Ricchini M, Rendina M (2020) Discovery of a potent
dual inhibitor of acetylcholinesterase and butyrylcholinesterase with antioxidant activity that
alleviates Alzheimer-like pathology in old APP/PS1 mice. J Med Chem 64(1):812–839
59. Czarnecka K, Girek M, Kr˛ecisz P, Skibiński R, Ł˛atka K, Jończyk J, Bajda M, Kabziński J,
Majsterek I, Szymczyk P, Szymański P (2019) Discovery of new cyclopentaquinoline analogues
as multifunctional agents for the treatment of Alzheimer’s disease. Int J Mol Sci 20(3):498
60. Maciejewska K, Czarnecka K, Kr˛ecisz P, Niedziałek D, Wieczorek G, Skibiński R, Szymański
P (2022) Novel cyclopentaquinoline and acridine analogs as multifunctional, potent drug
candidates in Alzheimer’s disease. Int J Mol Sci 23(11):5876
61. Safarizadeh H, Garkani-Nejad Z (2019) Molecular docking, molecular dynamics simulations
and QSAR studies on some of 2-arylethenylquinoline derivatives for inhibition of Alzheimer’s
amyloid-β aggregation: insight into mechanism of interactions and parameters for design of
new inhibitors. J Mol Graph Model 1(87):129–143
62. Jiang MY, Han C, Zhang C, Zhou Q, Zhang B, Le ML, Huang MX, Wu Y, Luo HB
(2021) Discovery of effective phosphodiesterase 2 inhibitors with antioxidant activities for
the treatment of Alzheimer’s disease. Bioorg Med Chem Lett 1(41):128016
63. Zhang C, Yang K, Yu S, Su J, Yuan S, Han J, Chen Y, Gu J, Zhou T, Bai R, Xie Y (2019) Design,
synthesis and biological evaluation of hydroxypyridinone-coumarin hybrids as multimodal
monoamine oxidase B inhibitors and iron chelates against Alzheimer’s disease. Eur J Med
Chem 15(180):367–382
64. Rullo M, Catto M, Carrieri A, de Candia M, Altomare CD, Pisani L (2019) Chasing ChEs-MAO
B multi-targeting 4-aminomethyl-7-benzyloxy-2H-chromen-2-ones. Molecules 24(24):4507
65. Palareti G, Legnani C, Cosmi B, Antonucci E, Erba N, Poli D, Testa S, Tosetto A (2016)
DULCIS (D-dimer-ultrasonography in combination italian study) investigators (see appendix),
De Micheli V, Ghirarduzzi A. Comparison between different D-D imer cutoff values to assess
the individual risk of recurrent venous thromboembolism: analysis of results obtained in the
DULCIS study. Int Lab Hematol 38(1):42–9
66. Sivakumar M, Saravanan K, Saravanan V, Sugarthi S, Kumar SM, Alhaji Isa M, Rajakumar P,
Aravindhan S (2020) Discovery of new potential triplet acting inhibitor for Alzheimer’s disease
via X-ray crystallography, molecular docking and molecular dynamics. J Biomol Struct Dyn
38(7):1903–1917
67. Sepehri N, Mohammadi-Khanaposhtani M, Asemanipoor N, Hosseini S, Biglar M, Larijani B,
Mahdavi M, Hamedifar H, Taslimi P, Sadeghian N, Gulcin I (2020) Synthesis, characterization,
molecular docking, and biological activities of coumarin–1, 2, 3-triazole-acetamide hybrid
derivatives. Arch Pharm 353(10):2000109
68. Baruah P, Rohman MA, Yesylevskyy SO, Mitra S (2019) Therapeutic potency of substituted
chromones as Alzheimer’s drug: Elucidation of acetylcholinesterase inhibitory activity through
spectroscopic and molecular modelling investigation. BioImpacts 9(2):79
69. Taslimi P, Turhan K, Türkan F, Karaman HS, Turgut Z, Gulcin I (2020) Cholinesterases, α-
glycosidase, and carbonic anhydrase inhibition properties of 1H-pyrazolo [1, 2-b] phthalazine-
5, 10-dione derivatives: Synthetic analogues for the treatment of Alzheimer’s disease and
diabetes mellitus. Bioorg Chem 1(97):103647
70. Zhou Y, Li J, Yuan H, Su R, Huang Y, Huang Y, Li Z, Wu Y, Luo H, Zhang C, Huang L
(2021) Design, synthesis, and evaluation of dihydropyranopyrazole derivatives as novel PDE2
inhibitors for the treatment of Alzheimer’s disease. Molecules 26(10):3034
71. Acar Cevik U, Saglik BN, Levent S, Osmaniye D, Kaya Cavuşoglu B, Ozkay Y, Kaplancikli ZA
(2019) Synthesis and AChE-inhibitory activity of new benzimidazole derivatives. Molecules
24(5):861
3 Role of Computational Modeling in Drug Discovery for Alzheimer’s … 107

72. Mehrazar M, Hassankalhori M, Toolabi M, Goli F, Moghimi S, Nadri H, Bukhari SN,

Firoozpour L, Foroumadi A (2020) Design and synthesis of benzodiazepine-1, 2, 3-triazole
hybrid derivatives as selective butyrylcholinesterase inhibitors. Mol Diversity 24:997–1013
73. Osmaniye D, Sağlık BN, Acar Çevik U, Levent S, Kaya Çavuşoğlu B, Özkay Y, Kaplancıklı ZA,
Turan G (2019) Synthesis and AChE inhibitory activity of novel thiazolylhydrazone derivatives.
Molecules 24(13):2392
74. Luo L, Song Q, Li Y, Cao Z, Qiang X, Tan Z, Deng Y (2020) Design, synthesis and evaluation
of phthalide alkyl tertiary amine derivatives as promising acetylcholinesterase inhibitors with
high potency and selectivity against Alzheimer’s disease. Bioorg Med Chem 28(8):115400
75. Lan JS, Zeng RF, Jiang XY, Hou JW, Liu Y, Hu ZH, Li HX, Li Y, Xie SS, Ding Y, Zhang T
(2020) Design, synthesis and evaluation of novel ferulic acid derivatives as multi-target-directed
ligands for the treatment of Alzheimer’s disease. Bioorg Chem 1(94):103413
76. Begum S, Nizami SS, Mahmood U, Masood S, Iftikhar S, Saied S (2018) In-vitro evaluation
and in-silico studies applied on newly synthesized amide derivatives of N-phthaloylglycine as
Butyrylcholinesterase (BChE) inhibitors. Comput Biol Chem 1(74):212–217
77. Xie SS, Liu J, Tang C, Pang C, Li Q, Qin Y, Nong X, Zhang Z, Guo J, Cheng M, Tang W (2020)
Design, synthesis and biological evaluation of rasagiline-clorgyline hybrids as novel dual
inhibitors of monoamine oxidase-B and amyloid-β aggregation against Alzheimer’s disease.
Eur J Med Chem 15(202):112475
78. Meden A, Knez D, Malikowska-Racia N, Brazzolotto X, Nachon F, Svete J, Sałat K, Grošelj U,
Gobec S (2020) Structure-activity relationship study of tryptophan-based butyrylcholinesterase
inhibitors. Eur J Med Chem 15(208):112766
79. Jiang X, Zhang Z, Zuo J, Wu C, Zha L, Xu Y, Wang S, Shi J, Liu XH, Zhang J, Tang W (2021)
Novel cannabidiol−carbamate hybrids as selective BuChE inhibitors: docking-based fragment
reassembly for the development of potential therapeutic agents against Alzheimer’s disease.
Eur J Med Chem 5(223):113735
80. Ali S, Asad MH, Khan F, Murtaza G, Rizvanov AA, Iqbal J, Babak B, Hussain I (2020)
Biological evaluation of newly synthesized biaryl guanidine derivatives to arrest β-secretase
enzymatic activity involved in Alzheimer’s disease. Biomed Res Int 11:2020
81. Das M, Prakash S, Nayak C, Thangavel N, Singh SK, Manisankar P, Devi KP (2018) Dihydroac-
tinidiolide, a natural product against Aβ25-35 induced toxicity in Neuro2a cells: synthesis, in
silico and in vitro studies. Bioorg Chem 1(81):340–349
82. Kumar A, Srivastava G, Negi AS, Sharma A (2019) Docking, molecular dynamics, binding
energy-MM-PBSA studies of naphthofuran derivatives to identify potential dual inhibitors
against BACE1 and GSK-3β. J Biomol Struct Dyn 37(2):275–290
83. Hu XM, Dong W, Cui ZW, Gao CZ, Yu ZJ, Yuan Q, Min ZL (2018) In silico identification of
AChE and PARP-1 dual-targeted inhibitors of Alzheimer’s disease. J Mol Model 24:1–9
84. Hassan M, Shahzadi S, Seo SY, Alashwal H, Zaki N, Moustafa AA (2018) Molecular
docking and dynamic simulation of AZD3293 and solanezumab effects against BACE1 to
treat Alzheimer’s disease. Front Comput Neurosci 1(12):34
Chapter 4
Computational Modeling
in the Development of Antiviral Agents

Priyank Purohit, Pobitra Borah, Sangeeta Hazarika, Gaurav Joshi,

and Pran Kishore Deb

Abstract As a result of the damage that viruses have done over time, humans
have developed a variety of defenses against viral illnesses, such as vaccines and
antiviral drugs for treatment. Since the 1950s, new viral illnesses including AIDS,
Hepatitis, and coronavirus infections like SARS, MERS, and COVID-19 have peri-
odically emerged, posing a challenge to the development of antiviral drugs. The
creation of computer models is an interactive, iterative process that blends empir-
ical datasets with known facts and assumptions (knowledge-driven or data-driven
approach). In order to allow system simulation, the generated models should ideally
offer reusability, composability, and interoperability. We surmise that the develop-
ment of computational and mathematical frameworks will not only assist the devel-
opment of newer antivirals, but simulating viral infections will also help in incor-
porating progressive immunosenescence and finding host genetic factors to expand
the knowledge of infectious disease to an unprecedented level of detail. In addition
to the fundamental molecular aspects of viral infection, this chapter emphasizes the
fundamentals of computer modeling and discusses the relationship between in silico
experiments and viral infections.

Keywords Virus–receptor interactions · Virtual screening · Molecular docking ·

In silico · Antiviral drug development

P. Purohit · P. Borah (B) · S. Hazarika · G. Joshi

School of Pharmacy, Graphic Era Hill University, Dehradun, Uttarakhand 248002, India
e-mail: [email protected]
S. Hazarika
Department of Pharmaceutical Engineering and Technology, Indian Institute of Technology
(Banaras Hindu University), Varanasi, Uttar Pradesh 221005, India
P. K. Deb
Department of Pharmaceutical Sciences, Faculty of Pharmacy, Philadelphia University, PO Box 1,
Amman 19392, Jordan

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 109
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_4
110 P. Purohit et al.

4.1 Introduction

In today’s scenario, it is still a matter of debate about how to define viruses.

Although many different names have been given to viruses over the years, including
“pathogenic entities”, “microscopic parasites”, “small-sized infectious agents”, “col-
lections of genes surrounded by a protein coat”, and “organic particles”, all of these
definitions are conclusive of the fact: viruses are non-living beings as they are depen-
dent on other cells for survival and replication. In spite of their shared need for
host cells for replication, viruses display a remarkable degree of diversity, with four
distinct types recognized based on their physical characteristics: endogenous viral
components, temperate viruses, virus-like particles, and virus particles or virions [1].
Of these, virus particles cause human infection. Nucleic acid genome and an inner
nucleic acid core containing genetic information like DNA or RNA are enclosed
and protected by a protein casing called a capsid. There are viruses that have an
additional coating of a lipid bilayer taken from host cell membranes. Most extracel-
lular enveloped virions include additional viral particles including surface glycopro-
teins and matrix proteins [2]. Multiple cell and virion surface factors regulate host
cell contact and intracellular viral tropism, facilitating effective viral propagation
and replication. Two major modes of viral tropism have been identified: receptor-
dependent tropism, in which invasion is initiated by the binding of virus to specific
molecules of host cells, and receptor-independent tropism, where a vector is needed
by virus to enter the host cell [3]. Most viruses employ receptor-dependent tropism,
meaning that the exclusive expression of a given receptor on a particular cell type is
predictive of that cell type’s susceptibility to viral infection, cell entrance, and viral
reproduction. Therefore, the viral structure can be divided based upon two main
functions: (1) the infectious function leading to viral entrance into the host cells by
causing identification of cellular receptors by surface proteins and (2) the protective
function carried out by capsid proteins [4]. The host immune system exerts selection
pressure on the viral surface proteins or protein domains. Antibody-binding areas,
also known as epitopes, undergo frequent mutation as a result of this selection pres-
sure resulting in the continual emergence of new viral entities or strains. Besides
mutation via selection, some viruses may result in novel viral subtypes with geneti-
cally recombined regions of nucleic acid. Viruses with greater adaptability are often
to blame for the widespread distribution of illness throughout vast areas or even
globally, leading to pandemics.
By analyzing the biological interactions of medications with viral molecular
components using available datasets, computational modeling offers a more effective
way to speed up drug discovery or repurposing of antiviral treatments. Target-based
and disease-based computational techniques are both used to identify drug molecules.
While the latter makes use of pre-existing databases to identify novel indications for
current medications [5] by comparing them with illness characteristics, the former
generates data on drug–target interaction. A fully regulated in silico reconstruction
of mechanisms and interaction processes is possible with computational modeling
and simulation [6]. Following that, experiments are carried out computationally in a
4 Computational Modeling in the Development of Antiviral Agents 111

“simulation”. Mathematical and computational modeling consequently addresses an

increasing number of crucial aspects of infection dynamics [7]. Being able to deter-
mine previously unknown infection parameters makes this particularly intriguing. It
is now beyond dispute that CADD plays a critical role in the drug design process and
that, over time, it can both speed up the discovery of viable therapeutic candidates and
lower the associated costs. CADD and its approaches can have an impact on different
stages of the drug discovery pipeline by using knowledge and information about
biological targets or knowledge about a ligand with demonstrated bioactivity. Inves-
tigations into the mechanism of drug action are made possible by CADD methods’
capacity to illuminate drug–target interactions at the atomistic level, revealing atom-
istic insights that have an impact on drug development. Since CADD approaches
also enable modeling of complicated biological processes that, up until now, seemed
impossible to investigate using experiments, they strive to complement rather than
replace established in vitro and in vivo experimental procedures. We speculated that
the creation of mathematical frameworks for modeling viral infections will aid in
integrating progressive immunosenescence and identifying host genetic influences
to enhance the understanding of infectious disease to an unheard-of level of detail.
The basics of computer modeling are highlighted in this chapter, and we go into
greater detail on relating in silico experiments to viral infections.

4.2 Brief History and Structure of Viruses

Viruses are the smallest (~20–300 nm), non-cellular, obligate intracellular infectious
agents that utilize the host cellular machinery for their survival and multiplication.
Among 400 viruses known to be pathogenic to human, the zoonoses that spread
from insects or animals to humans are particularly considered to be more dangerous
for the human race [8, 9]. Transmission of the viruses occurs by various routes,
including air or droplet infection (e.g., smallpox, influenza, measles, chicken pox,
mumps, viral pneumonia, rubella, coronavirus disease-2019, etc.), bite of arthropods
(e.g., Colorado tick fever and yellow fever); physical contact (e.g., genital herpes,
common cold, rabies, cold sores, acquired immunodeficiency syndrome or AIDS,
etc.); and water- or food-borne (e.g., hepatitis A, poliomyelitis, and viral gastroen-
teritis). Historically, viral infections led to millions of casualties, including epidemics
or pandemics starting from the smallpox epidemic (AD 165–180 and AD 251–266)
to Spanish flu (1918–1920), human immunodeficiency virus (HIV) infection (1981–
present), severe acute respiratory syndrome or SARS (2002–2003), Middle East
respiratory syndrome or MERS (2012–present), West African Ebola (2014–2016),
and the COVID-19 (2019–present) pandemic [10, 11]. Accordingly, the humans are
adopting different counter-measures over time to control the viral infections or their
spreads, which includes development of antiviral agents and vaccines. As the viruses
are dependent on host cell machinery, it is difficult to develop an effective and safer
antiviral drug without potentially harming the host [12].
112 P. Purohit et al.

The viruses can be classified as DNA viruses or RNA viruses simply because
they either contain DNA or RNA but not both. The majority of RNA viruses have
a single-stranded RNA (ssRNA) molecule, although a few may contain a double-
stranded RNA (dsRNA) molecule. If the RNA base sequence is identical to viral
mRNA, it is referred to as the positive (+) strand, whereas if it is complementary,
it is referred to as the negative (−) strand. In contrast, the majority of DNA viruses
possess dsDNA, while only a few include ssDNA [6]. A comprehensive taxonomy
of viruses is available at https://talk.ictvonline.org/. Based on their morphological
architecture, viruses can have helical symmetry, icosahedral (cubic) symmetry, or a
complex structure. For instance, viruses like influenza virus demonstrate standard
helical symmetry with viral genome bound to protein subunits (nucleocapsid helix)
encased in a lipid envelope, whereas human papillomavirus exhibits icosahedral
symmetry as they have 20 equilateral triangular faces in the capsid with 12 vertices.
Contrary to this, viruses (e.g., poxvirus) that lack helical or icosahedral symmetry
may exhibit more complicated forms such as an ovoid, spherical, or brick-shaped
pattern [13].
Apart from these, viruses can be further classified based on various considera-
tions, such as the type of genome they contain (DNA versus RNA, single strand
versus double strand, sense versus antisense, and linear versus circular), physico-
chemical properties, replication mechanism, and so on. The genetic blueprints that
a virus uses to replicate and achieve host lethality may vary slightly between copies.
This is especially true for retroviruses and some other RNA viruses [9]. In spite of
destroying the host cell and its machinery, certain viruses alter the cellular physi-
ology. Unchecked cellular proliferation is attributed to several infections that may
later progress into malignancy. Hepatitis B and C are two viruses that can infect
the liver and lead to chronic liver disease. For some people, the effects of chronic
hepatitis can last for years or even decades. In certain populations, the mild forms
of chronic hepatitis never progress to liver failure; however, some individuals may
eventually develop cirrhosis and liver failure. Typically, viruses target a single cell
type. For example, the common cold virus only infects cells in the upper respiratory
tract. Though viruses specifically infect plants and animals, some of them are only
transmitted by human beings [10].

4.3 Mechanism of Viral Infections

As an obligate intracellular parasite, viruses are absolutely dependent on the host

resources for their growth and replication processes. In a nutshell, the replication
cycle of a virus starts with the attachment phase, in which the virus attaches to the
host cell receptor primarily through electrostatic adsorption in a fortuitous manner.
For instance, HIV binds the CD4 receptor, rhinoviruses recognize the ICAM-1
receptor, and Epstein-Barr virus targets the CD21 receptor [14]. A study reported
that certain mammalian cells inhibit viral entry in non-infected cells by producing
proteins capable of restricting the binding of virus to the receptor [15]. Next, in
4 Computational Modeling in the Development of Antiviral Agents 113

the penetration phase, viruses invade the cell by various mechanisms, viz. envelope/
membrane fusion, receptor-mediated endocytosis, or through direct penetration of
membrane. It is followed by the viral genome (DNA or RNA) release in the cell
and exhibits an eclipse period, where the virus exploits host machinery to make
necessary viral proteins and undergoes replication. Permissive cells promote viral
replication to favor new virion production. Mostly, the host metabolic processes are
directed toward the generation of viral products, which ultimately lead to host cell
destruction. However, if the cell metabolism remains intact, the infected cell tends to
survive. Abortive infections may be observed in cases of defective virus invasion or
non-permissive cell infections. This results in latent infection, which can later cause
malignancy [14, 16].
The replication processes vary depending on the type of virus; however, tran-
scription followed by protein translation remains the eventual goal [17]. Typically,
the viruses containing dsDNA (except for poxvirus) require the host enzyme DNA-
dependent RNA polymerase for viral protein production, including essential enzymes
like DNA-dependent DNA polymerase. On the other hand, the hepatitis B virus
entails DNA polymerase to repair the DNA damages before transcription [18].
Contrariwise, an ssDNA virus initially needs to convert its genetic material into
dsDNA (sometimes with the assistance of a helper virus) prior to the translation of
mRNA. Notably, the RNA viruses have demonstrated evolution over time to obtain
different reproductive strategies compared to the DNA viruses. To understand the
ssRNA virus replication process, they may be grouped into three types. The first
group (e.g., togaviruses, picornaviruses, and flaviviruses) contains (+)-sense-strand
RNA, which is processed by the ribosome into polyproteins. Subsequent autocatal-
ysis of polyproteins by the proteases yields the viral proteins. Nevertheless, a few
viruses, such as togaviruses, possess only a segment of the RNA for translation.
The template sense strand is used to yield the antisense strand by using RNA poly-
merase. Later, the yielded antisense strands are modified into infectious and sense-
strand RNAs, which are packed inside virions for future transmission processes [14,
19]. On the other hand, another virus group, including paramyxoviruses, orthomyx-
oviruses, arenaviruses, and rhabdoviruses, contains antisense RNA that is required
for transcription. These viruses utilize RNA transcriptase (usually found within
the virion) to generate short sense-strand RNAs coding for essential viral enzymes
required for replication. Subsequently, these enzymes are utilized to produce full-
length sense RNAs and multiple antisense genome in the progeny [14, 20]. The
third group includes retroviruses with ssRNA found in dimeric forms of both sense
and antisense strands. The reverse transcriptase (RT), carried in the virion, converts
the retroviral RNA to dsDNA. Transcription results in complementary DNA strand
formation, while RNAse H and ribonucleases digest the original strand. Ensuring
this dsDNA synthesis follows, which is later incorporated into the host DNA via
integrase. The result of transcription is full-length RNA (progeny) that is packaged
as virion materials and polyproteins that are required for the generation of essential
proteins [21–23].
During replication, viral genome-encoded structural proteins such as the lipid
envelope that contains antigenic glycoproteins are also formed. Cellular egression
114 P. Purohit et al.

occurs either by destruction of the cell due to the assembly of corresponding virions
(as in case of non-enveloped viruses including reoviruses, picornaviruses, etc.) or by
budding and shedding extracellularly following virion insertion into the membrane
(mostly seen in enveloped viruses such as herpesviruses, togaviruses, coronaviruses,
retroviruses, hepadnaviruses, and antisense RNA viruses). Generally, the manifesta-
tion of symptoms (though they may sometimes remain asymptomatic) is considered
a consequence of rapid viral replication leading to host cell injuries. Overall, viral
infection and its pathogenesis remain dependent on viral entry, replication, transmis-
sion, cellular injury, immune responses, and viral shedding. A detailed insight into
viral diseases is available in existing literature [refer to 14, 16, 24, 25].

4.4 Computational Modeling in Viral Infections

Mathematics and logic form the basis of computational modeling of biological

phenomena. In case of viral infections, computational biology techniques allow
close observation of the molecular interactions in the form of simulation of the
biological membranes, protein folding, protein–ligand complexes, protein–protein
complexes, and molecular structures [26, 27]. For instance, application of Statis-
tical Mechanics and Molecular Dynamics methods allows simulation of different
protein folding conformations. These methods can be used to simulate conforma-
tional changes of viral restriction factors (e.g., BST-2, IFITM, APOBEC3G, and
SAMHDI) during the activation of innate immune systems associated with viral
infections such as Zika, HIV, Hepatitis C, Western Nile, and Influenza [28]. Simi-
larly, Molecular Modeling is utilized for the 3D representation of protein structures
via conformational analysis. Several other computer programs allow construction
and identification of small-molecule library by analyzing different protein databases
(e.g., Protein Data Bank or PDB) [29]. The computational protein–protein interac-
tion (PPI) prediction tools mainly utilize information such as protein structure, phys-
iochemical properties, sequence information, semantic analysis, and earlier inves-
tigated interactions between cellular proteins and viral proteins [30]. In spite of
high recommendation of classical machine learning tools in the prediction of PPI
(given the availability of plenty of known interactions), newer models are demanded
for the prediction of virus–host PPI. Instead of doing learning tasks on separate
domains, multitask learning exploits relationships among various domains and learns
the problem concurrently, which improves performance [31, 32].
Broadly, computational chemistry can be divided into Molecular Mechanics and
Quantum Mechanics methods. The former method is based on Newtonian physics and
can calculate very big molecules by viewing a molecule as a collection of spheres
held together by springs. As long as the atoms’ hybridization is maintained, the
parameters for the molecule’s atoms in this method are essentially constant across
varied structures [33]. On the other hand, delineating a linear combination of atomic
orbitals as molecular orbitals, the Quantum Mechanics approaches take into account
4 Computational Modeling in the Development of Antiviral Agents 115

the motion of electrons while assuming that the nucleus is fixed. They can be catego-
rized into ab initio or semi-empirical techniques. The ab initio approaches warranting
solutions for a broad array of molecules make approximations based on the laws of
quantum mechanics for directly solving Schrodinger’s equation, whereas empirical
parameters or pre-calculated parameters from Schrodinger’s equation form basis
of the development of semi-empirical approaches [33]. Application of molecular
modeling of the interaction between fusion protein of human Respiratory Syncytial
Virus (RSV) and inhibitory small molecules is an example of use of computational
modeling in viral infections. To predict novel host–virus protein interactions, many
computational techniques have been developed. In innovative host–virus interac-
tion predictions, various predictive models have been put forth depending on the
availability of interaction information.
The five axes of classification for computational models in virology are: (1)
discrete versus continuous; (2) hypothesis-driven versus data-driven; (3) stochastic
versus deterministic; (4) spatiotemporal versus temporal; and (5) white-box versus
black-box models. Atomistic simulations and individual-virion simulations modeling
atoms and virions, respectively, as discrete entities are two examples of discrete
models. In continuous models, no independent representation of the individual enti-
ties is made; rather, monitoring of their dispersion in time and/or space or density is
done. How a virion becomes infectious is a crucial subject in virus assembly that has
been addressed using continuous modeling approaches. In the case of HIV, whereby
virion infectivity is essentially gained by proteolytic maturation, a functional fusion
machinery in the infectious virions is proposed to be assembled by matrix (MA)
domain cleaving off from the envelope (ENV) domain including the viral glyco-
protein, and allowing ENV to loosen up with the spread of MA in the virion and
clustering the trimers [34]. Spatiotemporal models directly depict the spatial local-
ization or dispersion of viruses. As opposed to geographic localization, temporal
models follow the dynamic evolution of an aggregated variable, such as the total
virus load or infection multiplicity. In viral kinetics models, this is typically the case
[35]. On the other hand, in stochastic models, some events occur probabilistically,
making it impossible to anticipate how the infection state will develop over time, but
it is possible to assess the likelihood of various developments. For instance, a “white
box” model for the modeling of infection transmission in a tissue culture model has
been constructed that takes the local virion concentration into account [36]. On the
contrary to stochastic models, deterministic models assume that all outcomes are
known in advance; therefore, this type of model often necessitates thorough under-
standing of the underlying molecular mechanics. The extrapolation of deterministic
models to acto-myosin-based cell motility has been done for several features of actin
polymerization [37, 38]. If the interactions between the proteins that make up the
capsid are known in atomic detail, deterministic models might be used to simulate the
disruption of a non-enveloped virion. Such models might prove to be better predic-
tive tools; for instance, as reported in the case of adenovirus, whereby it can foretell
if during virion drifting motions on the cell surface, the mechanical forces acting on
virus particles are adequate for the partial breakage of the virion upon entry [39–41].
Models that are hypothesis-driven are created based on an anticipation or a hunch,
116 P. Purohit et al.

even in the absence of initial data. A hypothesis is first formulated, after which it
is formalized—for instance, using mathematical formulas, sets of rules, or chemical
reactions. The generated model is then put through a simulation to investigate its
behavior and test it against observations and known facts in an effort to disprove the
hypothesis. Instead of formalizing a hypothesis, models can be learned from data.
This is frequently helpful in the early exploratory stages of a project or when looking
for higher-order patterns in data that are hidden to the human eye. Although statistical
analytic techniques have long done this, current advances in artificial intelligence and
machine learning have given data-driven modeling a new level of sophistication. The
correlations must be learned using supervised machine learning, which needs a lot of
training data from well-known situations. It is therefore not unexpected that some of
the earliest applications in virology were finding connections between viral infection
state and host cell gene expression levels, for instance in hepatitis B virus infections
[42]. Likewise, white-box models only have one or a few unidentified parameters. The
most concrete proof that a mechanism is sufficient is a white-box model. It is strong
proof that the modeled mechanisms are adequate if, for instance, all binding affini-
ties, infection probabilities, and diffusion constants are measured independently and
the model accurately reproduces the data. Black box models, in contrast to white-
box models, are completely identified by parameter fitting. As a result, once the
fundamental mechanism is understood, they offer indirect methods of estimating not
directly measurable or observable quantities. They do, however, always leave some
room for uncertainty regarding the real mechanism because many processes may be
able to reproduce the same data for various parameter values [43]. The ability to
draw statistical conclusions from experimental data is a crucial function of simula-
tions. The Bayesian framework or the maximum-likelihood framework is commonly
used for inference. For both, an observation process “forward model” is necessary.
The mechanism of virus propagation seemingly marked by plaque dynamics can be
deduced using virus plaque formation in infected tissue simulations [36].

4.4.1 Virtual Screening (VS)

By using algorithms to sort through a library of tiny chemicals, VS, an in silico

approach, can identify those that have the potential ability to bind a target molecule.
In contrast to experimental approaches like High Throughput Screening, in which
numbers of compounds are experimentally evaluated, it has allowed scientists to
obtain lead compounds for binding a biological target from millions of molecules in
a dataset within a short span of time while being inexpensive. This is made feasible by
the ongoing development and improvement of more powerful computers and reliable
algorithms that steadily reduce the amount of time needed for such analyses while
enhancing the precision of the projected interactions [44, 45]. Two distinct computer-
aided drug design or CADD methodologies are structure-based drug design (SBDD)
and ligand-based drug design (LBDD) (Fig. 4.1). Using SBDD techniques, it is
4 Computational Modeling in the Development of Antiviral Agents 117

Computer-aided drug design

(CADD)

Structure-based drug design Ligand-based drug design

(SBDD) (LBDD)

Docking and Scoring Pharmacophore

QSAR
modelling

Virtual Screening (VS)

Compound selection/
lead optimization

In vitro assays

Fig. 4.1 Diagram depicting the workflow of virtual screening

possible to pinpoint key sites and interactions that are essential to each macromolec-
ular target’s specific biological activity. These macromolecular targets are typically
proteins or RNA. Once the biological target has been identified, information on
its structure must be gathered from a number of sources. The PDB is the most
significant of these, as it contains macromolecular structures that have been solved
experimentally by NMR, X-ray crystallography, or cryoelectron microscopy [33,
46]. The alternative, in the lack of a structure that can be solved experimentally,
is to obtain a model that has previously been proposed or to build one based on
homology using software like the resilient Modeler. The next step after obtaining the
biological target’s structure is to define the binding locations where the screening
compounds will be docked. When ligands are present on repositories, the binding
site may even be identified with structures if it is well-established. But research can
be done using tools like MDpocket [47], which can simultaneously find transitional
pockets and prospective target sites while investigating a new biological target or
looking at additional potential allosteric sites in the structure. The next stage is the
docking of the compounds in the biological target’s binding site after it has been
identified and clustered [48].
The structure–activity relationship (SAR)—a concept central to LBDD—is estab-
lished by focusing on known compounds or ligands in order to determine a correlation
between their physical–chemical characteristics and biological activity. The design
of novel medications or drug optimization can be influenced by this knowledge [49].
Quantitative structure–activity relationship (QSAR), a method of calculation, allows
for the quantification of molecular descriptors. Additionally, there is a chance for a
connection with research into the compounds’ tridimensional structure (3D-QSAR).
Many methodologies go beyond tridimensional descriptors and explore ligands in
a variety of other factors, such as solvation profile and various conformations. As
118 P. Purohit et al.

an illustration, consider 4D-QSAR, which represents the many conformations of the

same ligand as a new variable by investigating quantum mechanics (QM) or molecular
mechanics (MM). The descriptors in any SAR methodology must be carefully picked
depending on each inquiry, excluding highly correlated ones, in order to minimize
overlapping data that can be found by multiple linear regression (MLR) analysis or
even principal component analysis (PCA). [33]. Another analysis in LBDD suggests
using a pharmacophore model to uncover more potent ligands. The minimum elec-
trostatic energy and spatial distribution of some functional groups must be proposed
as being required for effective activity against the targeted target when modeling.
Functional groups in a known ligand or set of ligands, from hydrogen bond accep-
tors and donors to hydrophobic or aromatic groups, must be examined. Analyzing a
pharmacophore model, which is closely related to the 3D-QSAR method, is a highly
successful strategy to screen for new active chemicals as well as to improve the
known ones. Software that can be utilized in this technique includes LigandScout,
Discovery Studio Visualizer, and platforms for visualizing and analyzing structures.
The next steps are as follows after choosing the first strategy (either SBDD or
LBDD) to go through virtual screening. The preparation of the chemicals database
is the first and most important step. In particular, if the 3D geometry of such objects
is not present in the selected database, it entails finding and getting it. A variety of
programs, like Open Babel and LigPrep, can be used to transform 2D structures into
3D ones. The Cambridge Structure Database (CSD), which houses the crystal struc-
tures of compounds, can be used to obtain more reliable structures. Checking the
tautomeric and protonation state is also necessary before moving on to the following
steps to enable confidence and accurate docking interactions. Additionally, in terms
of the target preparation, in addition to correcting the structure for a specific pH, an
intriguing optional step can be to group the binding site among a collection of struc-
tures, either from various structures found in repositories or from MD simulations, in
order to find a docking pose that is more precise. As a result, clustering can be done
using MD software utilizing tools like GROMACS’s clustering function or Amber’s
CPPTRAJ [44]. In most cases, compounds under investigation are docked in a grid
assigned by one of the several docking programs (e.g., Glide and AutoDock4.2). In
the subsequent phase, the best candidate ligands will be ranked and selected using
the ratings based on electrostatic and Van der Waals interactions at the binding site
that are provided in this stage. By evaluating and identifying the substances that
can bind and those that do not interact in a stable manner at the biological binding
site, this crucial phase aids in the reduction of the number of molecules that need
to be studied [50]. The final stage before starting in vitro testing with the remaining
collection of drugs is the examination of pharmacokinetics properties, which follows
the energy of the complex protein–ligand [49]. One of the problems undermining
the study is inaccuracies in the energy of the systems. Deep learning and machine
learning are currently the most successful strategies created to address that. Machine
learning can use a training dataset and be applied to the test’s list of compounds, or
it can use the test’s list of compounds directly, in which case the algorithm will try
to find a pattern without any prior knowledge. Examples of algorithms that can be
used for this assessment include Naive Bayes (NB), K-Nearest Neighbor (KNN), and
4 Computational Modeling in the Development of Antiviral Agents 119

Random Forest (RF). Contrarily, deep learning is regarded as a subset of machine

learning techniques that utilizes neural networks. The issue is solved hierarchically
using layers of different nonlinear layouts of processing units. Due to the inherent
flexibility of its connections, it is a powerful technique to process many descriptors
quickly, but reproducibility improvements are still needed [51].

4.4.2 Molecular Docking

One of the first steps in evaluating compounds as possible drug molecules is molec-
ular docking, which works in tandem with VS and seeks to forecast the orientation
and conformation of ligands at the target binding site (Fig. 4.2) [33]. The orienta-
tion and conformation of a ligand at the receptor-binding site are analyzed using a
search approach in molecular docking, and the binding free energy, or affinity, of a
receptor-ligand complex is predicted using a scoring function. Molecular docking
requires the molecular structures of the macromolecule receptor and ligand. It is
necessary to first define the three-dimensional structure of the receptor, which is
frequently a protein. Ideally, this should be done using experimental data that may
be available in databases like the Protein Data Bank (PDB). PDB contains the atom
coordinates in individual molecules as well as complexes that have been resolved
using experimental techniques like X-ray diffraction and nuclear magnetic reso-
nance. Alternative computational methods can be employed for proteins that have
not been solved yet. One of these methods is called comparative modeling, in which
the amino acid sequence of the upcoming protein is compared to sequences of other
proteins with existing structures [52]. Theoretical computation can also be used to
undertake ab initio modeling for structure prediction [33]. The receptor-binding site
that needs to be examined must be chosen manually and defined by providing the
coordinates, or it must be chosen automatically using the coordinates of ligands that
have already been bonded. Prediction algorithms can be employed to locate potential
cavities in situations where the site is unknown. In order to reduce computing costs, a
grid was created, which may be conceptualized as a box made up of a cubic network
of points where the interaction energy is calculated. The grid delimits the binding
site for the typical molecular docking calculation. Following the precalculation of
the receptor potential energy for each grid point, the electrostatic and Lennard–Jones
energies of the ligand’s interaction with each grid point are computed [53]. Investi-
gating the protonation states of the amino acids that interact at the receptor-binding
site is also crucial. The structures of the ligands can be found in open databases like
Zinc and PubChem. Dedicated molecular drawing applications like MarvinSketch,
ChemDraw, and Avogadro can be used to create input files with three-dimensional
coordinates [54]. When creating the ligand file, some factors need to be taken into
consideration. Charges, stereochemical geometry, and the protonation states of chem-
ical groups are some of these [55]. In addition to the pH, the interaction with the
macromolecule binding site should be taken into account when determining the proto-
nation states of the ligand groups. To utilize Molecular Docking with the receptor,
120 P. Purohit et al.

Target protein Ligand Prediction & Geometry & Binding site Molecular docking
(Receptor) structure Modelling Protonation states detection

Search Score
algorithm functions

Structure of 3D coordinates Protein structure Convert 2D to 3D Identification of Conformational search

target proteins of ligands modelling based structure & determine favourable ligand and the selection of best
(3D coordinates) on homology protonation states of binding site pose by the scoring
the amino acids functions

Examples of software used

• PDB • ChemDraw • Modeller • Maestro • MolDock • AutoDock

• PubChem • Pymol • Lomets • AutoDock • FTMAP • Vina
• ZINC • Avogadro • I-TASSER • LigPrep • FragmentHot • GOLD
• ChEMBL • ChemSketch • Robetta • UCSF Chimera • SpotMap • DockThor
• PDBbind • MarvinSketch • Swiss-Model • MOE tool • DoGSiteScorer • Glide

Fig. 4.2 Examples of software used in different stages of molecular docking

software like LigPrep can produce representative structures of these modifications

[33]. The type of docking that needs to be computed must be specified because it has
a direct impact on the outcomes. The computations can take either a rigid or flex-
ible approach to the ligand and receptor-binding point. Due to the greater number of
degrees of freedom that must be considered, flexible receptors and ligands have calcu-
lations that are more complex than those for rigid ones. Some docking software for
flexible receptors and ligands is based on Koshland’s induced-fit hypothesis, which
states that both must tolerate slight changes in the binding that may take place [56].

4.4.3 Molecular Dynamics (MD)

The variation of atomic locations in a molecule as a function of time is described

by Molecular Dynamics (MD), a computational method using Newton’s laws. As
a result, it differs from the docking technique, which relies mostly on stochastic
processes. MD can be used to analyze traits including secondary structure content,
side chain orientation, loop conformation, and the energy of long-term interac-
tions with various molecules, like proteins and ligands. Because of this, it exhibits
results comparable to those of experimental techniques like Nuclear Magnetic Reso-
nance, but with the benefit of higher cost savings. Additionally, MD may be used to
larger molecular systems, has better biological system reproduction, and takes less
time [33]. MD simulation entails a number of processes, from getting the protein
structure to parameterization, equilibration, minimization, simulation, and analysis.
Molecular dynamics has several uses, including protein structure prediction, protein–
lipid, protein–protein, and protein–ligand interaction analyses, viral studies, and drug
4 Computational Modeling in the Development of Antiviral Agents 121

creation for the widest range of disorders. The most challenging aspect of MD, in
the opinion of many, is the analysis of MD simulations. The stability of variables
like pressure, temperature, density, volume, and total energy during the course of the
simulation needs to be examined first and foremost. You can proceed to the other
analyses if these were stable. The root mean square deviation, which assesses how
much the protein conformation changed throughout the course of the MD simula-
tion in contrast to the initial crystallographic structure, is one of the most frequently
performed analyses, particularly for proteins.
Apart from all these, numerous methods have been used in recent years to
find novel antiviral agents with improved resistance profiles and new scaffolds,
such as topology-matching design, targeted covalent inhibitors, proteolysis-targeting
chimera (PROTAC), and ribonuclease targeting chimera (RIBOTAC) [57]. The
“topology-matching design” theory for virus inhibitors was put to the test in 2020 by
Nie et al. [58]. They created a nano-inhibitor that had the same nano-topological struc-
ture as the IAV virions and had hetero-multivalent inhibitory effects on HA and NA.
The created nano-inhibitor was able to stop the viral particle’s attachment to host cells
and extracellular neutralization. The viral replication was significantly slowed down
by six orders of magnitude and was inhibited to a level of more than 99.99% even 24 h
after infection, indicating that a nano-inhibitor of this kind could be an effective anti-
influenza agent. A spiky nano-inhibitor with a topology similar to IAV virions was
also discovered by the researchers. The same group demonstrated topology-matched
hetero-multivalent nanostructures as wide-spectrum IAV inhibitors [59]. The rational
design of targeted covalent inhibitors has tremendously benefited from the advance-
ment of structural biology and bioinformatics. Covalent inhibitors can engage with
specific target proteins to produce covalent bonds that alter protein shape, impairing
the protein’s ability to function normally [60]. The process of creating a covalent
bond with the target can be broken down into two related but distinct steps: (1) the
inhibitor binds to the target reversibly, positioning the functional groups on weak
electrophilic ligands next to particular nucleophilic residues on the protein; and (2)
the ligand interacts with the functional groups involved in the protein to create the
covalent bond [61]. PROTACs have emerged as a new paradigm for drug develop-
ment to target proteins by encouraging and realizing the degradation of targets via
the Ubiquitin–proteasome system [62]. PROTACs are hetero-bifunctional molecules
made up of a linker, an E3 ubiquitin ligase recruiting ligand, and a ligand for the
protein of interest (POI). To reduce the distance between them in vivo, bifunctional
PROTAC molecules have one end that binds to the POI and the other end that attaches
to an E3 ligase. The POI is subsequently ubiquitylated by an E2 enzyme through the
E3 ligase, and the POI is then degraded by the proteasome [63, 64]. Antiviral agents
have recently been steadily discovered using this technology. A novel method for
RNA degradation is called Ribonuclease Targeting Chimera (RIBOTAC). In order
to destroy the viral genome, RIBOTAC has an RNA-binding small molecule and a
ribonuclease (RNase) L-recruiting module [65]. RNase L functions in innate immu-
nity and is present in all cells in trace amounts as an inactive monomer. During viral
infection, it becomes activated and dimerizes with innate substrate specificity [66].
122 P. Purohit et al.

To achieve the effect of selective cleavage, RIBOTACs locally recruit RNase L to

the anticipated target.

4.5 Virus-Surface Proteins and Receptor Interaction

Enveloped viruses employ fusion glycoproteins as an adornment for their surface,

and these proteins are crucial for the virus entry to the host cells. Fusion proteins are
major components found on the viral surface, playing a crucial role in the infection
process. As such, they are great targets for the development of antivirals and serve as
the main immunogens in a variety of vaccine types. The refolding of fusion proteins
from a highly strained and unstable prefusion conformation to a very stable postfu-
sion conformation is assumed to supply the free energy that catalyzes the joining of
host and virus membranes. The three main types of fusion proteins are Class I, Class
II, and Class III fusion proteins [67]. Even though they all fuse in a similar way, these
proteins have varied structures and respond to various stimuli. Class I proteins are
tetramers with a higher α-helix content; in their postfusion conformation, they show
a coiled-coil-helix encircled by three helices at the carboxy terminus (six α-helix
bundles). Class II fusion proteins have a pre- and postfusion structural signature of
β-sheets and are either homo- or heterodimers before fusion. Finally, class III fusion
proteins have a mixed α-helical and β-sheet structure in both their prefusion and
postfusion trimer stages. Class I and II fusion proteins’ activation procedures need
proteolytic processing of either the fusion protein or its partner protein; however,
class III fusion proteins could not require processing to allow for cell entrance. Once
fusion-competent, low pH is the main catalyst for class II fusion proteins, whereas
class III fusion proteins are activated by interactions between a partner protein and
a host cell receptor or by low pH. Class III fusion proteins may achieve an equi-
librium between the prefusion and postfusion states, allowing the transition to be
reversible, in contrast to the vast majority of viral fusion proteins that experience
irreversible conformational changes upon activation (see Fig. 4.3) [68]. Class I fusion
proteins are essential components of the fusion machinery of many different virus
families, including the Orthomyxoviridae, Paramyxoviridae, Coronaviridae, Retro-
viridae, and Filoviridae. SARS-CoV-2 is only one example of a virus that has a class
I fusion protein, but this kind of fusion protein has been extensively investigated
since it is so common. In the context of the COVID-19, it is crucial to explore class
I fusion proteins as targets of therapeutic importance. Class I fusion proteins, as an
overarching example, are thought to be constructed from single-chain precursors
that undergo proteolytic maturation to acquire fusion competence throughout their
development [69]. A metastable structure that defines the prefusion state transitions
through a number of intermediate phases after being activated to reach the postfu-
sion state. The transition to a low pH environment and interactions with cell surface
coreceptors (e.g., the envelope protein of HIV interacting with the CCR5, or inter-
actions with cell surface receptors coupled with localized protease cleavage (e.g.,
the SARS-CoV-2 S protein binding to ACE2 receptor and is subsequently cleaved)
4 Computational Modeling in the Development of Antiviral Agents 123

Fully
developed
hairpin
state
Host cell

Funsion of host
membrane
and Viral membrane

Funsion
Peptide

Viral membrane

Prefusion state Pre-hairpin state Post fusion state

Fig. 4.3 General mechanism of virus–host cell interaction

are examples of triggers [70]. By stabilizing any of the conformations before the
postfusion state is reached, the fusion process may be stopped. Inhibiting receptor
binding is another option for avoiding receptor attachment altogether, and neutral-
izing antibodies are used for both of these purposes. For HA, for example, bnAbs
target a conserved portion (the receptor-binding site pocket) in an otherwise variable
region (the rest of the HA molecule). However, the antibodies’ ability to neutralize a
wide range of HA variants is hampered by the significant diversity around this site.
Because of the high degree of conservation of the HA stalk, anti-stem antibodies tend
to offer highly broad neutralizing potential because they target a particular epitope
proximal to the fusion loop [71].
Numerous studies have examined the importance of certain cellular receptors as
tropism-determining factors. The early studies on picornaviruses revealed a relation-
ship between receptors on cells and the well-known in vivo tropism of poliovirus
using organ minces and homogenates. Polioviruses could be absorbed by intestine
and central nervous system (CNS) tissues from both humans and monkeys, but not
by human lung, heart, or skin tissues. However, because receptors were also discov-
ered on the monkey heart, human liver, and skeletal muscle, the connection was not
strong [72]. In addition, poliovirus vaccine strains that do not cause cell damage have
been demonstrated to attach to brain regions. Another piece of evidence linking cell
surface receptors and pathogenicity is the similarity between the categorization of
viruses into subgroups based on patterns of disease and the grouping of viruses by
receptor specificities. Finally, picornaviruses’ growth specificities have been shown
using organ cultures. Some types of rhinoviruses, for instance, can only replicate
in trachea organ cultures. Coxsackie viruses A1 and A5 can develop in primary,
124 P. Purohit et al.

differentiating mouse muscle cultures but not in non-differentiating mouse cultures.

Receptors for human enteroviruses, on the other hand, have been found in non-host
species and on tissues unrelated to the virus’ pathogenesis [73].
The virulence and cell tropism of the reovirus serotypes 1 and 3 have been genet-
ically characterized to differ from one another. Reovirus serotype 3 and clone 1 have
been shown to have different central nervous system cell tropisms depending on the
presence of the S1 gene, which codes for the hemagglutinin. In newborn mice, HA3
(which contains nine genes from type I and the gene encoding the hemagglutinin
from type 3), causes a fatal encephalitis with neuronal destruction but no ependymal
cell damage. The differing cell-type selectivity of the serotypes can be explained by
the viral hemagglutinin’s apparent interaction with receptors on the surface of either
neuronal or ependymal cells. Reovirus type 1 and clone 3.HA1 binding to isolated
human cells served as an in vitro confirmation of these findings [74].
The M form of the encephalomyocarditis (EMC) virus induces a diabetes-like
disease in some mouse strains by infecting and destroying pancreatic β cells. Since
cultured pancreatic beta cells from resistant strains of mice are less able to absorb
infectious EMC virus than beta cells from strains that are susceptible to EMC-induced
diabetes, it appears that genetically determined variations in the surface viral recep-
tors on these cells may be one of the factors regulating susceptibility to disease [75].
Lymphocytes may have a more targeted immune response to some viruses if
they include viral receptors. Measles virus has a receptor on T cells and measles
infection is linked to decreased tuberculin skin hypersensitivity and decreased helper
cell activity. As with humans (T8 + cells), the Ly2,3 fraction of murine T cells
is a primary target for reovirus type 3. The viral hemagglutinin is responsible for
this binding. Additionally, when examined in vitro, suppressor T cells produced by
reovirus type 3 may prevent Con A proliferation. It appears that reovirus type 3
produces functionally active suppressor T cells in vitro before engaging the viral
hemagglutinin with a specific receptor on the Ly2,3 subset of mice lymphocyte [76].

4.6 Antivirals Targeting Viral Surface Proteins

By inhibiting viral replication in host cells, several small-molecule antiviral

medicines decrease viral load, illness severity, or death from viral infections. Current
antiviral medications, such as those used to combat HIV (abacavir; Ziagen), HSV
(famciclovir; Famvir), and HBV (tenofovir; Rescriptor), are a prime example of this
phenomenon (lamivudine; Epivir). For respiratory viruses like influenza, RSV, or
coronaviruses, however, there are very limited treatment choices [77]. Oseltamivir,
zanamivir, and peramivir, three antivirals, have an additional inhibitory mechanism
that blocks the release of viral offspring from infected cells by inhibiting the activity
of the neuraminidase protein. They do not involve the fusion proteins themselves but
instead interfere with the molecular pathway by binding to a surface glycoprotein.
But they are only useful in the early stages of an infection [78].
4 Computational Modeling in the Development of Antiviral Agents 125

Supportive therapy (including acetaminophen for pyrexia and i.v. fluids for dehy-
dration) and neutralizing antibodies (e.g., palivizumab) are often used to treat RSV
infections. It is important to note that the cost of antibody-based therapy may add
up rapidly. There are now four small-molecule antivirals licensed by the FDA for
treatment against influenza viruses. Among them are medications that are effec-
tive against influenza A and B viruses and that target the influenza neuraminidase
protein, such as oseltamivir phosphate, zanamivir, and peramivir. The issue is that
these drugs frequently only function when used promptly after infection. Despite the
fact that small-molecule drugs and/or their combinations have been demonstrated to
be effective, the research and development of innovative small molecules against an
emerging pandemic like COVID-19 has proven to be challenging [79]. Challenges
include extensive development times, limited specificity, and typically poor affini-
ties (given the tiny surface area of small compounds) that may lead to unwanted side
effects or the requirement for repeated dosing to reach positive therapeutic objec-
tives. Already licensed medications may target additional dangerous viruses, making
drug repurposing a possible option to generating novel small molecules for treating
viral infections. Remdesivir, Gilead Sciences, USA, was in this position; it had been
designed for Ebolavirus and now has emergency FDA permission for COVID-19.
Also being evaluated for treatment against SARS-CoV-2 infections is the favipiravir,
which was first authorized for use against influenza in Japan. Both compounds target
the viral RNA-dependent RNA polymerase in their triphosphate forms after being
phosphorylated in vivo as prodrugs (RdRp). To prevent viral replication and spread,
bioactive triphosphates compete with purine nucleosides for incorporation into viral
RNA.
Alternative medicines for treating viral infections include those based on proteins.
Higher specificity and affinity than small-molecule medicines are made possible by
the vast protein–protein and/or protein-glycan interfaces. Monoclonal antibodies
are the most widely used protein-based therapies, although they may be expen-
sive to produce and may not be very stable (bacterial diseases, viral infections,
arthritis, cancer, etc.). Antibody-based treatments have also proven an effective
strategy in the ongoing COVID-19 epidemic. In particular, two treatment options
have been approved for emergency use by the FDA to protect against SARS-CoV-2:
the Regeneron cocktail (REGN-COVTM), casirivimab along with imdevimab, and
bamlanivimab combined with etesevimab, which target the receptor-binding domain
of the S protein at different sites. This development happened incredibly fast, with
the pandemic starting in February 2020, antibodies being found in May 2020, clin-
ical trials starting in November 2020, and FDA approval coming in February 2021,
respectively [80].
Using tiny protein-based inhibitors is an alternate approach to developing big
antibody molecules. Tiny, extremely stable, and easy-to-produce proteins may be
developed to block functional locations in a target biomolecule, similar to naturally
occurring inhibitors where small proteins attach to bigger proteins to occlude their
action (say cystatin in trypsin). Recent research has shown that computational design
may create very potent protein–protein inhibitors. The capacity of the influenza HA
protein to facilitate fusion or attach to human cell receptors has been blocked, for
126 P. Purohit et al.

instance, by inhibitors built from scratch to target-specific sections of the protein.

Inhibitors of both classes were able to prevent and treat influenza in mice. Highly
stable picomolar inhibitors were created lately by blocking the receptor-binding
site of the S protein with small protein inhibitors [81]. These fine-tuned molecules
prevented viral replication for three days and reversed infection for twenty-one days
in Golden Syrian hamsters. The low molecular weight (<5 kDa) of this antibody
compared to the typical 150 kDa of a monoclonal antibody means that far less
material has to be generated to have the same efficiency, opening up a fantastic
possibility to provide for a greater number of individuals. The use of computational
design has several benefits over more conventional approaches, one of which is the
robust binding of these tiny proteins [82]. Computational protein design, in contrast
to the traditional method of identifying monoclonal antibodies from a pool, gives the
designer full authority over the inhibitor’s target area and allows for the optimization
of the interaction of interest. Last but not least, tiny protein inhibitors are a potential
method to combating viral infections since their great stability might circumvent
storage and transportation hurdles that often impede the supply of medications to
distant places [83].

4.7 Applications of Computational Modeling in Antiviral

Drug Discovery

The study of viruses is an intriguing application of molecular dynamics. The satellite

tobacco mosaic virus was the first fully atomized non-enveloped virus that Schulten
and colleagues described in 2006 [84]. In terms of enveloped viruses, Ayton and Voth
performed the first simulation of capsid of an immature HIV virus in 2010 [85]. The
majority of investigations, however, focus on simulations of viral capsids and proteins
since viral particles are too big for all-atom simulations. The study of more intricate
mechanisms such replication, membrane attachment and fusion, viral maturation,
and nuclear entry represents a constraint for this field. However, the study of viral
proteins yields a wealth of crucial knowledge that also enables MD to develop drugs
for viral diseases [86, 87]. Studying the connections between the Ebola virus’ VP24
protein and the human protein KPNA5, Pappalardo and colleagues determined that
this protein is essential for the virus’ pathogenicity [88]. Molecular dynamics was
used by Nasution and colleagues to undertake a drug design study that resulted in the
identification of potential molecules [89]. Another example is Zhang and Zheng’s
study of the influenza B virus, which provided structural and dynamic details of
the effect of serine triad on proton conduction in the tetrameric influenza B channel
M2, essential to the virus life cycle [90]. To uncover possible inhibitors of the Zika
NS2B-NS3 utilizing a collection of more than 7 million chemicals from the ZINC15
database, for instance, Bowen and colleagues used consensus-based docking, MD
modeling, and binding energy estimates [91]. Although there is a long history of
mathematical modeling of viral infections and immunity, the idea of quantifying the
4 Computational Modeling in the Development of Antiviral Agents 127

pathogen’s within-host dynamics only became widely used in clinical practice after
the publication of two seminal papers in 1995 [92, 93]. These papers used modeling
to demonstrate that HIV infection was a highly dynamic process, in which the virus
was rapidly replicating and being cleared within infected individuals. Due to the
fact that HIV infection takes over ten years to progress to AIDS and because for the
most of this time plasma HIV levels remain relatively stable, many people previously
believed that HIV was a “dormant” virus comparable to other lentiviruses. When it
was discovered that HIV replicated quickly (and erratically), it became clear that
medication resistance was unavoidable if only one or two drugs were employed, for
which the virus may develop resistance with one or two point mutations. As a result,
it was determined that combination therapy involving at least three medications was
required. The longevity of infected cells, the effectiveness of drug treatment, the
modes of action of treatment, the dynamics of diverse populations of infected cells,
etc., were all clarified by modeling HIV and then other infections.
To confirm the efficacy of drug combinations for epidemiological outbreaks in
sizable populations, machine learning techniques can be used to create complex
network models. In order to halt the spread of AIDS in the U.S. epidemic networks,
Herrera and his colleagues proposed the first artificial neural network (ANN) model
for the prediction of HAART cocktails in 2015. They did this by compiling datasets
from ChEMBL (anti-HIV chemical compounds), the AIDSVu database (HIV surveil-
lance reports), and the Census Bureau (socioeconomic data) [94]. In silico fragment-
based drug design was carried out for drug–molecule interaction research, the first
multitasking model for quantitative structure-biological effect relationships (mtk-
QSBERs) was predicted, and its molecular components were virtually screened for
the Hepatitis C virus infection [95]. Data on AIDS in the United States has been
linked to ChEMBL data using ANNs. Network prediction models called ANNs
are mostly employed in the creation of pharmaceuticals and medicinal chemistry.
Using drug-drug similarity complex networks, Gonzalez-D’az et al. predicted novel
antimicrobial medicines and targets using the MARkovian CHemicals IN SIlico
DEsign (MARCH-INSIDE) method [96]. Novel stochastic moments are generated
by computational analyses of structural stability connections [97]. In order to simu-
late their effects on the relationships between protein structure stability, Markovian
Backbone Negentropies (MBNs) have been introduced. It has also been reported on
an MBN for the computational investigation of structure/stability relationships based
on a Markov chain model of electron delocalization along the protein backbone [98].
These methods primarily concentrate on cheminformatic methods for drug design,
drug development, and drug–target interactions in in silico clinical fields, particularly
with regard to viral illnesses.
The orthomyxoviridae family of RNA viruses is responsible for the annual
seasonal pandemic of acute respiratory illness known as influenza. To combat
influenza, pharmaceutical companies have focused on inhibiting neuraminidase
(NA), the enzyme responsible for viral entry into the mucous membranes of the
host. With this in mind, VS has been utilized effectively to discover new inhibitors
of this target. They described a search of the National Cancer Institute (NCI)
compound database using shape and chemical features to identify potential analogues
128 P. Purohit et al.

of katsumadain A, a proven NA inhibitor. Within the micromolar range, five of the

NA-inhibiting drugs tested (including four flavonoids) showed significant activity
against three oseltamivir-susceptible H1N1 strains [99]. A strategy for forecasting
influenza’s dominant hemagglutinin gene, Plotkin et al. [100] proposed a virus and
investigated the evolution of antigenic over the host genes. Clustering analysis was
performed on the spatiotemporal viral swarm dispersal and the evolution of hemag-
glutinin structure. To cluster the viral sequences, a critical length scale in amino acid
space was lastly used.
To combat HIV retrovirus, scientists have taken several different overarching
strategies, including (1) preventing the retrovirus’ genetic material from replicating
or being inserted within the host cell, (2) preventing the virus from entering the
cell, and (3) inhibiting the virus from fully assembling and maturing. In the first
VS run, the known RT inhibitor dihydroxy benzoyl naphthyl hydrazone was used
as the query molecule in a shape-based screening of the NCI database. The most
active hits from this process were used for a second VS using a combination ligand-
based technique that includes 2D and 3D similarity searches, and ligand-based phar-
macophore screening. In micromolar concentrations, some compounds identified
through new scaffolds exhibited strong anti-RT action, blocking both RT functions
and RT-associated ribonuclease H [101]. Finally, the capsid protein has been identi-
fied as a promising target for the design of next-generation antiretroviral medicines
because of its central involvement in HIV-1 assembly and maturation. The Curelli
group performed a VS using docking and then looked for analogues. The C-terminal
section of the HIV-1 capsid is the target of small-molecule inhibitors, and these
have been discovered. In infected cells, the most effective chemicals significantly
decreased infectivity [102]. Short linear motifs-based prediction methods for PPIs
between HIV-1 and host proteins have been proposed by Becerra et al. [103]. They
used three filtering techniques to find groups of linear motifs: (1) those conserved
in viral proteins (C); (2) those found in disordered areas (D); and (3) those that are
uncommon or sparse in a set of randomly generated viral sequences (R). To locate
the disordered protein areas among the HIV-1 sequences and host sequences, these
three sets have finally been used. Their research demonstrates that the majority of the
virus’s conserved linear motifs are found in disorganized areas. A group proposed a
way to simulate the network of virus–human interactions using motif–domain inter-
actions in [104]. A bi-clustering strategy was put forth by Mukhopadhyay et al. [105]
to forecast HIV-1-infected proteins using interaction-based analysis. By using the bi-
clustering technique, a set of association rules was extracted from the interaction of
HIV-1 proteins. In order to forecast novel interactions between human proteins, a
set of high-confidence rules was ultimately extracted. This same group [106] made
further advancements to their work by adding type and direction (virus-to-host and
host-to-virus)-based bi-clustering to already-known interactions to forecast novel
host proteins.
An approach to anticipate the interactions between human proteins and HIV-1
based on protein structural similarity was proposed by Doolittle and group [107]. In
this method, two crystal structures are compared to determine the structural similarity
of host to that of pathogen proteins. It is assumed that targets will be identified for
4 Computational Modeling in the Development of Antiviral Agents 129

human proteins that share a lot of structural similarities with HIV proteins and their
known interaction partners. They used a similar methodology to create a network of
interactions between the host and the Dengue virus [108].
Several effective docking studies of the E-protein have been published. Two hit
compounds demonstrated micromolar antiviral activity against dengue virus, while
one molecule displayed antiviral activity against West Nile virus and yellow fever
virus, which is an intriguing finding. Through docking of over a million in-house
compounds, the authors found a molecule that suppresses dengue fusion in the low
micromolar range. To combat HCV and other members of the Flaviviridae family,
NS5 polymerase has emerged as a promising drug development target [109].
Although popular methods based on structure-based approaches, such as molec-
ular docking, will always be helpful, they become computationally taxing when
used to find pan-antivirals via multitarget inhibition against many viral proteins. One
strategy focuses on models that combine perturbation theory and machine learning
(PTML), which, by using Box-Jenkins operators, can combine in vitro screening
data that includes various viral proteins from different viruses [110]. This makes it
possible to examine FDA-approved medications and experimentally validate those
that are projected to be versatile inhibitors of the viral proteins as prospective thera-
pies for SARS-CoV-2 and other viral infections. In the second method, medicines are
virtually screened using PTML models in inhibition/replication experiments against
various viruses. Once more, in biological testing against SARS-CoV-2, the medica-
tions projected to be pan-antivirals will take precedence. Modeling anti-HIV action
through multitarget inhibition [111], connecting in vitro activity with epidemiolog-
ical consequences [96], and virtually screening antiretroviral drugs [112] are all
successful uses of PTML models in antiviral research. PTML models could thus be
simply modified and created for the two aforementioned scenarios. Alignment-free
multitarget (AFMT) model development is the third strategy [113]. These compu-
tational methods are created by calculating molecular descriptors from the amino
acid sequences of the proteins and the 2D structures of the ligands/inhibitors. The
advantage of the AFMT models over their PTML equivalents is that the former can
forecast the inhibitory potency of huge libraries of (either virtual or already synthe-
sized) compounds against proteins other than those utilized to construct the AFMT
models.
Drugs may be designed and engineered using computational approaches. Compu-
tational approaches are convenient for high throughput screening of existing medi-
cations due to their short time requirements, which may be used to find treatments
for new ailments and to foresee their side effects. In a process known as drug repur-
posing, it is highly desirable to identify substances that have already been tested and
approved for use on the market and that exhibit activity against the new issue in order
to combat disease outbreaks or the ongoing emergence of drug-resistant diseases.
The reason this is crucial is that when repurposing an already-commercial compound,
several experimental steps can be skipped because information on their pharmacoki-
netic, pharmacodynamics, and toxicity profiles has been established, and they are,
in some cases, safe for human administration [114]. Given that creating a new treat-
ment is tedious and expensive before it can be made available to the public, this is
130 P. Purohit et al.

particularly interesting and useful in the recent Zika, Ebola, and COVID-19 virus
epidemics. Following the conventional steps in a VS process, the repurposing process
starts evaluating databases of already-commercial chemicals like ZINC. Strong hard-
ware and algorithms are consequently required for this fight since in an outbreak
event urgency is needed to stop the disease’s progress and the deaths that follow.
The most recent instance of this is the COVID-19 pandemic, which tested scientists’
abilities to the limit in order to find the quickest solutions to stop its spread, i.e.,
by developing, in a short period of time, the screening over thousand hits among
1.3 billion already regulated compounds against the virus main protease [114–116].
For instance, computational researches have identified Lopinavir and Ritonavir to be
potential inhibitors against SARS-CoV-2 main protease (Mpro) [117, 118].

4.8 Conclusion

In virology, computational techniques have advanced significantly. They cover a

wide range of machine learning techniques for automated model extraction, now-
standard computational image analysis techniques, and identifying probable inter-
actions between virions and host cell compartments. Since many of these advance-
ments were initially motivated by virology-related applications, the project was truly
interdisciplinary. In silico reconstitution is made possible by computer simulations of
learned or postulated models, which can also demonstrate the sufficiency of a process
rather than just a collection of constituent parts or molecules. Different resolutions
of simulations are possible, ranging from continuum models to all-atom molecular
dynamics simulations. In addition, we anticipate that a new generation of AI-driven
protein modeling tools, e.g., AlphaFold, will offer even greater advancements in
protein models for novel viruses. Nevertheless, when characterizing viral proteins,
de novo modeling should be utilized with caution and supported by tests because
their extremely wide structural repertoire might not be captured during training of an
AI system. Furthermore, it is extremely difficult to characterize the structural proper-
ties of the macromolecular complexes generated by viral proteins. The next step is to
create brand-new techniques for precise de novo protein complex characterization,
similar to AI-driven techniques for protein structure prediction. We anticipate that
computational modeling will continue to play a significant role in global viral disease
response, treatment, and prevention given the historical contributions of mathemat-
ical and computer modeling to understanding the spread of infectious illnesses and
their consequences in humans.

Conflict of Interest None to declare.

4 Computational Modeling in the Development of Antiviral Agents 131

References

1. Narkhede YB, Gonzalez KJ, Strauch E-M (2021) Targeting viral surface proteins through
structure-based design. Viruses 13:1320. https://doi.org/10.3390/v13071320
2. Lanjanian H, Nematzadeh S, Hosseini S et al (2021) High-throughput analysis of the interac-
tions between viral proteins and host cell RNAs. Comput Biol Med 135:104611. https://doi.
org/10.1016/j.compbiomed.2021.104611
3. Tellinghuisen TL, Rice CM (2002) Interaction between hepatitis C virus proteins and host cell
factors. Curr Opin Microbiol 5:419–427. https://doi.org/10.1016/S1369-5274(02)00341-7
4. Carvalho FA, Carneiro FA, Martins IC et al (2012) Dengue virus capsid protein binding to
hepatic lipid droplets (LD) is potassium ion dependent and is mediated by LD surface proteins.
J Virol 86:2096–2108. https://doi.org/10.1128/JVI.06796-11
5. Kumar R, Harilal S, Al-Sehemi AG et al (2021) The chronicle of COVID-19 and possible
strategies to curb the pandemic. Curr Med Chem 28:2852–2886. https://doi.org/10.2174/092
9867327666200702151018
6. Ji Z, Yan K, Li W et al (2017) Mathematical and computational modeling in complex biological
systems. Biomed Res Int 2017:e5958321. https://doi.org/10.1155/2017/5958321
7. Brodland GW (2015) How computational models can help unlock biological systems. Semin
Cell Dev Biol 47–48:62–73. https://doi.org/10.1016/j.semcdb.2015.07.001
8. Woolhouse MEJ, Gowtage-Sequeria S (2005) Host range and emerging and reemerging
pathogens. Emerg Infect Dis 11:1842–1847. https://doi.org/10.3201/eid1112.050997
9. Weiss RA, McMichael AJ (2004) Social and environmental risk factors in the emergence of
infectious diseases. Nat Med 10:S70-76. https://doi.org/10.1038/nm1150
10. Huremović D (2019) Brief history of pandemics (pandemics throughout history). Psychiat
Pandemics 7–35. https://doi.org/10.1007/978-3-030-15346-5_2
11. Piret J, Boivin G (2021) Pandemics throughout history. Front Microbiol 11. https://doi.org/
10.3389/fmicb.2020.631736
12. He H (2013) Vaccines and antiviral agents. In: Current issues in molecular virology-viral
genetics and biotechnological applications. IntechOpen
13. Fermin G (2018) Virion structure, genome organization, and taxonomy of viruses. viruses
17–54. https://doi.org/10.1016/B978-0-12-811257-1.00002-4
14. Lemke TL, Williams DA, Roche VF, Zito SW (2012) Foye’s principles of medicinal chemistry,
7th edn. Wolters Kluwer Health Adis (ESP)
15. Ganser-Pornillos BK, Chandrasekaran V, Pornillos O et al (2011) Hexagonal assembly of a
restricting TRIM5alpha protein. Proc Natl Acad Sci U S A 108:534–539. https://doi.org/10.
1073/pnas.1013426108
16. Patrick GL (2013) An introduction to medicinal chemistry. Oxford University Press
17. Novoa RR, Calderita G, Arranz R et al (2005) Virus factories: associations of cell organelles
for viral replication and morphogenesis. Biol Cell 97:147–172. https://doi.org/10.1042/BC2
0040058
18. Lu X, Block T (2004) Study of the early steps of the Hepatitis B Virus life cycle. Int J Med
Sci 1:21–33
19. Mukhopadhyay S, Kuhn RJ, Rossmann MG (2005) A structural perspective of the flavivirus
life cycle. Nat Rev Microbiol 3:13–22. https://doi.org/10.1038/nrmicro1067
20. Sedlmeier R, Neubert WJ (1998) The replicative complex of paramyxoviruses: structure and
function. In: Maramorosch K, Murphy FA, Shatkin AJ (eds) Advances in virus research.
Academic Press, pp 101–139
21. Rosenberg ZF, Fauci AS (1991) Immunopathogenesis of HIV infection. FASEB J 5:2382–
2390. https://doi.org/10.1096/fasebj.5.10.1676689
22. Basavapathruni A, Anderson KS (2007) Reverse transcription of the HIV-1 pandemic. FASEB
J 21:3795–3808. https://doi.org/10.1096/fj.07-8697rev
23. Goodman LS (2001) Goodman & Gilman’s the pharmacological basis of therapeutics.
McGraw-Hill, New York, pp 27–2141
132 P. Purohit et al.

24. De Clercq E, Li G (2016) Approved antiviral drugs over the past 50 years. Clin Microbiol
Rev 29:695–747
25. De Clercq E (2004) Antivirals and antiviral strategies. Nat Rev Microbiol 2:704–720
26. Halder AK, Dutta P, Kundu M et al (2018) Review of computational methods for virus–host
protein interaction prediction: a case study on novel Ebola–human interactions. Brief Funct
Genomics 17:381–391. https://doi.org/10.1093/bfgp/elx026
27. Jenner AL, Aogo RA, Davis CL et al (2020) Leveraging computational modeling to understand
infectious diseases. Curr Pathobiol Rep 8:149–161. https://doi.org/10.1007/s40139-020-002
13-x
28. Silva T dos SC da (2019) Inibição da replicação do influenza através da modulação de fatores
restritivos pelos ligantes dos receptores CCR5 e CXCR4
29. Hagen JB (2000) The origins of bioinformatics. Nat Rev Genet 1:231–236. https://doi.org/
10.1038/35042090
30. Zheng L-L, Li C, Ping J et al (2014) The domain landscape of virus-host interactomes. Biomed
Res Int 2014:867235. https://doi.org/10.1155/2014/867235
31. Kshirsagar M, Murugesan K, Carbonell JG, Klein-Seetharaman J (2017) Multitask matrix
completion for learning protein interactions across diseases. J Comput Biol 24:501–514.
https://doi.org/10.1089/cmb.2016.0201
32. Kshirsagar M, Carbonell J, Klein-Seetharaman J (2012) Techniques to cope with missing data
in host–pathogen protein interaction prediction. Bioinformatics 28:i466–i472. https://doi.org/
10.1093/bioinformatics/bts375
33. de Queiroz Simões RS, Ferreira MS, Dumas de Paula N et al (2020) Computational modeling
in virus infections and virtual screening, docking, and molecular dynamics in drug design.
Netw Syst Biol. https://doi.org/10.1007/978-3-030-51862-2_12
34. Chojnacki J, Staudt T, Glass B et al (2012) Maturation-dependent HIV-1 surface protein
redistribution revealed by fluorescence nanoscopy. Science 338:524–528. https://doi.org/10.
1126/science.1226359
35. Srivastava R, You L, Summers J, Yin J (2002) Stochastic versus deterministic modeling of
intracellular viral kinetics. J Theor Biol 218:309–321. https://doi.org/10.1006/jtbi.2002.3078
36. Yakimovich A, Gumpert H, Burckhardt CJ et al (2012) Cell-free transmission of human
adenovirus by passive mass transfer in cell culture simulated in a computer model. J Virol
86:10123–10137. https://doi.org/10.1128/JVI.01102-12
37. Labouesse C, Gabella C, Meister J-J et al (2016) Microsurgery-aided in-situ force probing
reveals extensibility and viscoelastic properties of individual stress fibers. Sci Rep 6:23722.
https://doi.org/10.1038/srep23722
38. Pollard TD, Cooper JA (2009) Actin, a central player in cell shape and movement. Science
326:1208–1212. https://doi.org/10.1126/science.1175862
39. Greber UF (2016) Virus and host mechanics support membrane penetration and cell entry. J
Virol 90:3802–3805. https://doi.org/10.1128/JVI.02568-15
40. Nakano MY, Boucke K, Suomalainen M et al (2000) The first step of adenovirus type 2
disassembly occurs at the cell surface, independently of endocytosis and escape to the cytosol.
J Virol 74:7085–7095
41. Burckhardt CJ, Suomalainen M, Schoenenberger P et al (2011) Drifting motions of the
adenovirus receptor CAR and immobile integrins initiate virus uncoating and membrane
lytic protein exposure. Cell Host Microbe 10:105–117. https://doi.org/10.1016/j.chom.2011.
07.006
42. Iizuka N, Oka M, Yamada-Okabe H et al (2002) Comparison of gene expression profiles
between hepatitis B virus- and hepatitis C virus-infected hepatocellular carcinoma by
oligonucleotide microarray data on the basis of a supervised learning method. Cancer Res
62:3939–3944
43. Sbalzarini IF, Greber UF (2018) How computational models enable mechanistic insights into
virus infection. Methods Mol Biol 1836:609–631. https://doi.org/10.1007/978-1-4939-8678-
1_30
4 Computational Modeling in the Development of Antiviral Agents 133

44. Sliwoski G, Kothiwale S, Meiler J, Lowe EW (2014) Computational methods in drug

discovery. Pharmacol Rev 66:334–395. https://doi.org/10.1124/pr.112.007336
45. D’Souza S, Prema KV, Balaji S (2020) Machine learning models for drug–target interactions:
current knowledge and future directions. Drug Discovery Today 25:748–756. https://doi.org/
10.1016/j.drudis.2020.03.003
46. Muratov EN, Amaro R, Andrade CH et al (2021) A critical overview of computational
approaches employed for COVID-19 drug discovery. Chem Soc Rev 50:9121–9151. https://
doi.org/10.1039/D0CS01065K
47. Schmidtke P, Bidon-Chanal A, Luque FJ, Barril X (2011) MDpocket: open-source cavity
detection and characterization on molecular dynamics trajectories. Bioinformatics 27:3276–
3285. https://doi.org/10.1093/bioinformatics/btr550
48. Anderson AC (2003) The process of structure-based drug design. Chem Biol 10:787–797.
https://doi.org/10.1016/j.chembiol.2003.09.002
49. Yu W, MacKerell AD (2017) Computer-aided drug design methods. Methods Mol Biol
1520:85–106. https://doi.org/10.1007/978-1-4939-6634-9_5
50. Kitchen DB, Decornez H, Furr JR, Bajorath J (2004) Docking and scoring in virtual screening
for drug discovery: methods and applications. Nat Rev Drug Discov 3:935–949. https://doi.
org/10.1038/nrd1549
51. Wang D, Cui C, Ding X, et al (2019) Improving the virtual screening ability of target-specific
scoring functions using deep learning methods. Front Pharmacol 10
52. Webb B, Sali A (2016) Comparative protein structure modeling using MODELLER. Curr
Protoc Protein Sci 86:2.9.1–2.9.37. https://doi.org/10.1002/cpps.20
53. Torres PHM, Sodero ACR, Jofily P, Silva-Jr FP (2019) Key topics in molecular docking for
drug design. Int J Mol Sci 20:E4574. https://doi.org/10.3390/ijms20184574
54. Hanwell MD, Curtis DE, Lonie DC et al (2012) Avogadro: an advanced semantic chemical
editor, visualization, and analysis platform. Journal of Cheminformatics 4:17. https://doi.org/
10.1186/1758-2946-4-17
55. Lousa D, Baptista AM, Soares CM (2012) Analyzing the molecular basis of enzyme stability in
ethanol/water mixtures using molecular dynamics simulations. J Chem Inf Model 52:465–473.
https://doi.org/10.1021/ci200455z
56. Pagadala NS, Syed K, Tuszynski J (2017) Software for molecular docking: a review. Biophys
Rev 9:91–102. https://doi.org/10.1007/s12551-016-0247-1
57. Xu S, Ding D, Zhang X, et al (2022) Newly emerging strategies in antiviral drug discovery:
dedicated to Prof. Dr. Erik De Clercq on occasion of his 80th anniversary. Molecules 27:850.
https://doi.org/10.3390/molecules27030850
58. Nie C, Parshad B, Bhatia S et al (2020) Topology-matching design of an influenza-neutralizing
spiky nanoparticle-based inhibitor with a dual mode of action. Angew Chem Int Ed 59:15532–
15536. https://doi.org/10.1002/anie.202004832
59. Nie C, Stadtmüller M, Parshad B, et al (2021) Heteromultivalent topology-matched nanos-
tructures as potent and broad-spectrum influenza A virus inhibitors. Sci Adv 7:eabd3803.
https://doi.org/10.1126/sciadv.abd3803
60. Bauer RA (2015) Covalent inhibitors in drug discovery: from accidental discoveries to avoided
liabilities and designed therapies. Drug Discov Today 20:1061–1073. https://doi.org/10.1016/
j.drudis.2015.05.005
61. Gehringer M, Laufer SA (2019) Emerging and re-emerging warheads for targeted covalent
inhibitors: applications in medicinal chemistry and chemical biology. J Med Chem 62:5673–
5724. https://doi.org/10.1021/acs.jmedchem.8b01153
62. Lai AC, Crews CM (2017) Induced protein degradation: an emerging drug discovery paradigm.
Nat Rev Drug Discov 16:101–114. https://doi.org/10.1038/nrd.2016.211
63. Paiva S-L, Crews CM (2019) Targeted protein degradation: elements of PROTAC design.
Curr Opin Chem Biol 50:111–119. https://doi.org/10.1016/j.cbpa.2019.02.022
64. Li X, Song Y (2020) Proteolysis-targeting chimera (PROTAC) for targeted protein degradation
and cancer therapy. J Hematol Oncol 13:50. https://doi.org/10.1186/s13045-020-00885-3
134 P. Purohit et al.

65. Costales MG, Aikawa H, Li Y et al (2020) Small-molecule targeted recruitment of a nuclease

to cleave an oncogenic RNA in a mouse model of metastatic cancer. Proc Natl Acad Sci U S
A 117:2406–2411. https://doi.org/10.1073/pnas.1914286117
66. Costales MG, Matsumoto Y, Velagapudi SP, Disney MD (2018) Small molecule targeted
recruitment of a nuclease to RNA. J Am Chem Soc 140:6741–6744. https://doi.org/10.1021/
jacs.8b01233
67. Sanders RW, Moore JP (2021) Virus vaccines: proteins prefer prolines. Cell Host Microbe
29:327–333. https://doi.org/10.1016/j.chom.2021.02.002
68. Castaño N, Cordts SC, Kurosu Jalil M et al (2021) Fomite transmission, physicochemical
origin of virus-surface interactions, and disinfection strategies for enveloped viruses with
applications to SARS-CoV-2. ACS Omega 6:6509–6527. https://doi.org/10.1021/acsomega.
0c06335
69. Churin Y, Roderfeld M, Roeb E (2015) Hepatitis B virus large surface protein: function and
fame. Hepatobiliary Surg Nutr 4:1–10. https://doi.org/10.3978/j.issn.2304-3881.2014.12.08
70. Hinuma S, Fujita K, Kuroda S (2021) Binding of nanoparticles harboring recombinant large
surface protein of hepatitis B virus to scavenger receptor class B type 1. Viruses 13:1334.
https://doi.org/10.3390/v13071334
71. Taha BA, Al Mashhadany Y, Bachok NN et al (2021) Detection of COVID-19 virus on surfaces
using photonics: challenges and perspectives. Diagnostics (Basel) 11:1119. https://doi.org/
10.3390/diagnostics11061119
72. Brown NA, Schrevens S, van Dijck P, Goldman GH (2018) Fungal G-protein-coupled recep-
tors: mediators of pathogenesis and targets for disease control. Nat Microbiol 3:402–414.
https://doi.org/10.1038/s41564-018-0127-5
73. Haverkamp A-K, Lehmbecker A, Spitzbarth I et al (2018) Experimental infection of
dromedaries with Middle East respiratory syndrome-Coronavirus is accompanied by massive
ciliary loss and depletion of the cell surface receptor dipeptidyl peptidase 4. Sci Rep 8:9778.
https://doi.org/10.1038/s41598-018-28109-2
74. Ram S, Gulati S, Lewis LA et al (2018) A novel sialylation site on neisseria gonorrhoeae
lipooligosaccharide links heptose II lactose expression with pathogenicity. Infect Immun
86:e00285-e318. https://doi.org/10.1128/IAI.00285-18
75. Maginnis MS (2018) Virus-receptor interactions: the key to cellular invasion. J Mol Biol
430:2590–2611. https://doi.org/10.1016/j.jmb.2018.06.024
76. Simone D, Al Mossawi MH, Bowness P (2018) Progress in our understanding of the pathogen-
esis of ankylosing spondylitis. Rheumatology (Oxford) 57:vi4–vi9. https://doi.org/10.1093/
rheumatology/key001
77. Mazzon M, Ortega-Prieto AM, Imrie D et al (2019) Identification of Broad-spectrum antiviral
compounds by targeting viral entry. Viruses 11:E176. https://doi.org/10.3390/v11020176
78. Villalón-Letelier F, Brooks AG, Londrigan SL. Reading PC MARCH8 restricts influenza A
virus infectivity but does not downregulate viral glycoprotein expression at the surface of
infected cells. mBio 12:e01484–21. https://doi.org/10.1128/mBio.01484-21
79. Krug RM, Aramini JM (2009) Emerging antiviral targets for influenza A virus. Trends
Pharmacol Sci 30:269–277. https://doi.org/10.1016/j.tips.2009.03.002
80. Vigant F, Santos NC, Lee B (2015) Broad-spectrum antivirals against viral fusion. Nat Rev
Microbiol 13:426–437. https://doi.org/10.1038/nrmicro3475
81. Meagher JL, Takata M, Gonçalves-Carneiro D et al (2019) Structure of the zinc-finger antiviral
protein in complex with RNA reveals a mechanism for selective targeting of CG-rich viral
sequences. Proc Natl Acad Sci U S A 116:24303–24309. https://doi.org/10.1073/pnas.191
3232116
82. Chakravarty M, Vora A (2021) Nanotechnology-based antiviral therapeutics. Drug Deliv
Transl Res 11:748–787. https://doi.org/10.1007/s13346-020-00818-0
83. Du L, Yang Y, Zhou Y et al (2017) MERS-CoV spike protein: a key target for antivirals.
Expert Opin Ther Targets 21:131–143. https://doi.org/10.1080/14728222.2017.1271415
84. Freddolino PL, Arkhipov AS, Larson SB et al (2006) Molecular dynamics simulations of the
complete satellite tobacco mosaic virus. Structure 14:437–449. https://doi.org/10.1016/j.str.
2005.11.014
4 Computational Modeling in the Development of Antiviral Agents 135

85. Ayton GS, Voth GA (2010) Multiscale computer simulation of the immature HIV-1 virion.
Biophys J 99:2757–2765. https://doi.org/10.1016/j.bpj.2010.08.018
86. Huber RG, Marzinek JK, Holdbrook DA, Bond PJ (2017) Multiscale molecular dynamics
simulation approaches to the structure and dynamics of viruses. Prog Biophys Mol Biol
128:121–132. https://doi.org/10.1016/j.pbiomolbio.2016.09.010
87. Perilla JR, Goh BC, Cassidy CK et al (2015) Molecular dynamics simulations of large macro-
molecular complexes. Curr Opin Struct Biol 31:64–74. https://doi.org/10.1016/j.sbi.2015.
03.007
88. Pappalardo M, Collu F, Macpherson J et al (2017) Investigating Ebola virus pathogenicity
using molecular dynamics. BMC Genomics 18:566. https://doi.org/10.1186/s12864-017-
3912-2
89. Nasution MAF, Toepak EP, Alkaff AH, Tambunan USF (2018) Flexible docking-based molec-
ular dynamics simulation of natural product compounds and Ebola virus nucleocapsid (EBOV
NP): a computational approach to discover new drug for combating Ebola. BMC Bioinf
19:419. https://doi.org/10.1186/s12859-018-2387-8
90. Zhang Y, Zheng Q-C (2019) What are the effects of the serine triad on proton conduction
of an influenza B M2 channel? An investigation by molecular dynamics simulations. Phys
Chem Chem Phys 21:8820–8826. https://doi.org/10.1039/C9CP00612E
91. Bowen LR, Li DJ, Nola DT et al (2019) Identification of potential Zika virus NS2B-NS3
protease inhibitors via docking, molecular dynamics and consensus scoring-based virtual
screening. J Mol Model 25:194. https://doi.org/10.1007/s00894-019-4076-6
92. Wei X, Ghosh SK, Taylor ME et al (1995) Viral dynamics in human immunodeficiency virus
type 1 infection. Nature 373:117–122. https://doi.org/10.1038/373117a0
93. Ho DD, Neumann AU, Perelson AS et al (1995) Rapid turnover of plasma virions and CD4
lymphocytes in HIV-1 infection. Nature 373:123–126. https://doi.org/10.1038/373123a0
94. Herrera-Ibatá DM, Pazos A, Orbegozo-Medina RA et al (2015) Mapping chemical structure-
activity information of HAART-drug cocktails over complex networks of AIDS epidemiology
and socioeconomic data of U.S. counties. Biosystems 132–133:20–34. https://doi.org/10.
1016/j.biosystems.2015.04.007
95. Speck-Planche A, Dias Soeiro Cordeiro MN (2017) Speeding up early drug discovery in
antiviral research: a fragment-based in silico approach for the design of virtual anti-hepatitis
C leads. ACS Comb Sci 19:501–512. https://doi.org/10.1021/acscombsci.7b00039
96. González-Díaz H, Herrera-Ibatá DM, Duardo-Sánchez A et al (2014) ANN multiscale model
of anti-HIV drugs activity versus AIDS prevalence in the US at county level based on informa-
tion indices of molecular graphs and social networks. J Chem Inf Model 54:744–755. https:/
/doi.org/10.1021/ci400716y
97. González-Díaz H, Prado-Prado F, Ubeira FM (2008) Predicting antimicrobial drugs and targets
with the MARCH-INSIDE approach. Curr Top Med Chem 8:1676–1690. https://doi.org/10.
2174/156802608786786543
98. Ramos de Armas R, González Díaz H, Molina R, Uriarte E (2004) Markovian Backbone
Negentropies: molecular descriptors for protein research. I. Predicting protein stability in Arc
repressor mutants. Proteins 56:715–723. https://doi.org/10.1002/prot.20159
99. Ghosh S, Nie A, An J, Huang Z (2006) Structure-based virtual screening of chemical libraries
for drug discovery. Curr Opin Chem Biol 10:194–202. https://doi.org/10.1016/j.cbpa.2006.
04.002
100. Plotkin JB, Dushoff J, Levin SA (2002) Hemagglutinin sequence clusters and the antigenic
evolution of influenza A virus. Proc Natl Acad Sci U S A 99:6263–6268. https://doi.org/10.
1073/pnas.082110799
101. Chowdhury KH, Chowdhury MR, Mahmud S et al (2020) Drug Repurposing approach against
novel coronavirus disease (COVID-19) through virtual screening targeting SARS-CoV-2 main
protease. Biology (Basel) 10:2. https://doi.org/10.3390/biology10010002
102. Jang WD, Jeon S, Kim S, Lee SY (2021) Drugs repurposed for COVID-19 by virtual screening
of 6218 drugs and cell-based assay. Proc Natl Acad Sci 118:e2024302118. https://doi.org/10.
1073/pnas.2024302118
136 P. Purohit et al.

103. Becerra A, Bucheli VA, Moreno PA (2017) Prediction of virus-host protein-protein interac-
tions mediated by short linear motifs. BMC Bioinformatics 18:163. https://doi.org/10.1186/
s12859-017-1570-7
104. Segura-Cabrera A, García-Pérez CA, Guo X, Rodríguez-Pérez MA (2013) A viral-human
interactome based on structural motif-domain interactions captures the human infectome.
PLoS ONE 8:e71526. https://doi.org/10.1371/journal.pone.0071526
105. Mukhopadhyay A, Maulik U, Bandyopadhyay S (2012) A novel biclustering approach to asso-
ciation rule mining for predicting HIV-1–human protein interactions. PLoS ONE 7:e32289.
https://doi.org/10.1371/journal.pone.0032289
106. Mukhopadhyay A, Ray S, Maulik U (2014) Incorporating the type and direction informa-
tion in predicting novel regulatory interactions between HIV-1 and human proteins using a
biclustering approach. BMC Bioinformatics 15:26. https://doi.org/10.1186/1471-2105-15-26
107. Doolittle JM, Gomez SM (2010) Structural similarity-based predictions of protein interac-
tions between HIV-1 and Homo sapiens. Virology Journal 7:82. https://doi.org/10.1186/1743-
422X-7-82
108. Doolittle JM, Gomez SM (2011) Mapping protein interactions between dengue virus and its
human and insect hosts. PLoS Negl Trop Dis 5:e954. https://doi.org/10.1371/journal.pntd.
0000954
109. Santos FRS, Nunes DAF, Lima WG et al (2020) Identification of Zika virus NS2B-NS3
protease inhibitors by structure-based virtual screening and drug repurposing approaches. J
Chem Inf Model 60:731–737. https://doi.org/10.1021/acs.jcim.9b00933
110. Nocedo-Mena D, Cornelio C, Camacho-Corona MDR et al (2019) Modeling antibacterial
activity with machine learning and fusion of chemical structure information with microor-
ganism metabolic networks. J Chem Inf Model 59:1109–1120. https://doi.org/10.1021/acs.
jcim.9b00034
111. Speck-Planche A, Kleandrova VV, Luan F, Cordeiro MNDS (2012) A ligand-based approach
for the in silico discovery of multi-target inhibitors for proteins associated with HIV infection.
Mol BioSyst 8:2188–2196. https://doi.org/10.1039/C2MB25093D
112. Vásquez-Domínguez E, Armijos-Jaramillo VD, Tejera E, González-Díaz H (2019) Multi-
output perturbation-theory machine learning (PTML) Model of ChEMBL data for antiretro-
viral compounds. Mol Pharm 16:4200–4212. https://doi.org/10.1021/acs.molpharmaceut.9b0
0538
113. Viña D, Uriarte E, Orallo F, González-Díaz H (2009) Alignment-free prediction of a drug-
target complex network based on parameters of drug connectivity and protein sequence of
receptors. Mol Pharm 6:825–835. https://doi.org/10.1021/mp800102c
114. Zheng W, Sun W, Simeonov A (2018) Drug repurposing screens and synergistic drug-
combinations for infectious diseases. Br J Pharmacol 175:181–191. https://doi.org/10.1111/
bph.13895
115. Schuler J, Hudson ML, Schwartz D, Samudrala R (2017) A systematic review of compu-
tational drug discovery, development, and repurposing for Ebola virus disease treatment.
Molecules 22:E1777. https://doi.org/10.3390/molecules22101777
116. Ton A-T, Gentile F, Hsing M et al (2020) Rapid identification of potential inhibitors of SARS-
CoV-2 main protease by deep docking of 1.3 billion compounds. Mol Inform 39:e2000028.
https://doi.org/10.1002/minf.202000028
117. Awad IE, Abu-Saleh AA-AA, Sharma S et al. High-throughput virtual screening of drug
databanks for potential inhibitors of SARS-CoV-2 spike glycoprotein. J Biomol Struct Dyn
1–14. https://doi.org/10.1080/07391102.2020.1835721
118. Mahdian S, Zarrabi M, Panahi Y, Dabbagh S (2021) Repurposing FDA-approved drugs to
fight COVID-19 using in silico methods: Targeting SARS-CoV-2 RdRp enzyme and host
cell receptors (ACE2, CD147) through virtual screening and molecular dynamic simulations.
Inform Med Unlocked 23:100541. https://doi.org/10.1016/j.imu.2021.100541
Chapter 5
Targeted Computational Approaches
to Identify Potential Inhibitors for Nipah
Virus

Sakshi Gautam and Manoj Kumar

Abstract Nipah virus (NiV) is a bat-borne, highly pathogenic RNA virus belonging
to the family Paramyxoviridae. It is known for causing lethal encephalitis in humans
with a high fatality rate. With time, the world has faced numerous outbreaks in various
regions such as Malaysia, Bangladesh, Philippines, and India. In this chapter, we have
summarized experimentally tested antivirals and computational approaches to predict
potential inhibitors against NiV. Various studies have been conducted in vitro and
in vivo to find the potent novel molecules or repurposed drugs. This section describes
the drug’s primary function in case of repurposing, targets, screening method, inhi-
bition efficiency, etc., from important studies. The computational section describes
the approaches to identifying the potent inhibitors against NiV. These approaches
include machine learning and QSAR-based prediction, molecular docking, molec-
ular dynamics, integrated structure- and network-based approach, Drug–target–
drug network-based approach, etc. In conclusion, this work will be helpful for the
researchers in examining antivirals against NiV.

Keywords Nipah virus · Antivirals · Inhibitors · Computational · Prediction ·

Repurposed drugs

5.1 Introduction

Paramyxoviruses such as Nipah virus (NiV) and Hendra virus (HeV) are life-
threatening human pathogens that cause a range of clinical manifestations from
asymptomatic to symptomatic infections such as acute respiratory illness and severe

S. Gautam · M. Kumar
Virology Unit and Bioinformatics Centre, Institute of Microbial Technology, Council of Scientific
and Industrial Research (CSIR), Sector 39A, Chandigarh 160036, India
S. Gautam · M. Kumar (B)
Academy of Scientific and Innovative Research (AcSIR), Ghaziabad 201002, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 137
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_5
138 S. Gautam and M. Kumar

acute encephalitis [1]. The mortality rate of NiV infection is approximately 40%–
75%, which varies depending on the clinical management and epidemiological
surveillance of the outbreak location (https://www.who.int/news-room/fact-sheets/
detail/nipah-virus). NiV infection in humans was first reported in September of the
year 1998 in Malaysia [2]. NiV virus was isolated from the cerebrospinal fluid (CSF)
of a patient in the Sungai Nipah village in 1999, therefore named this virus as “Nipah”.
From 2001 to 2013, various NiV outbreaks have occurred in Bangladesh during the
winter season in the 20 districts. A small outbreak of NiV also took place in the Philip-
pines in 2014. Various NiV outbreaks have occurred in India but not as frequently
as in Bangladesh. A large outbreak has taken place in Siliguri, West Bengal (2001),
and a small outbreak in Nadia district, West Bengal (2007) [3]. In May 2018, a NiV
outbreak also occurred in Kozhikode and Malappuram districts, Kerala [4, 5]. Till
now, the mortality rates of all the outbreaks are up to 91% including the outbreak in
Kerala in 2018 [6]. The natural reservoir of the NiV is the fruit bats; transmission
occurs from bats to humans, bats to pigs, pigs to humans, date palm sap to humans,
and between humans, bats, pigs, etc. [7].
NiV genome is a non-segmented negative sense, single-stranded RNA (ssRNA)
having a length of 18.2 kb, part of the genus Henipavirus, family Paramyxoviridae,
subfamily Paramyxovirinae, and order Mononegavirales [8]. It is categorized as
a Biosafety Level-4 (BSL-4) pathogen due to the high pathogenicity and lack of
efficacious therapeutics or vaccines. It encodes six structural proteins, i.e., fusion
protein (F), attachment glycoprotein (G), nucleoprotein (N), large protein or RNA
polymerase protein (L), phosphoprotein (P), and matrix protein. P gene encodes four
non-structural proteins, i.e., V, W, C, and P proteins [9, 10].
The incubation period for the NiV varies from 4 days to 2 weeks. Then, the
primary clinical non-specific symptoms start appearing such as fever, headache,
vomiting, dizziness, segmental myoclonus, hypertension and tachycardia, hypotonia
and areflexia, and abnormal pupillary reflexes. In severe cases, patients start devel-
oping symptoms such as pulmonary syndrome, diffuse alveolar shadowing and acute
respiratory distress syndrome (ARDS), motor deficits, reduced level of consciousness
[11, 12]
NiV virus emerged in 1988, and various outbreaks have occurred with very high
mortality rates over the last two decades. As it is a BSL-4 pathogen, highly pathogenic
in nature, minimal studies are carried out with a live virus, and no licensed thera-
peutics or vaccines are available to treat NiV disease. Till now, no antiviral is under
clinical trial for the treatment of NiV infection. So, there is an urgent need to develop
strategies to deal with NiV outbreaks. Various studies have been carried out to iden-
tify the potent inhibitors in vitro and in vivo against NiV. In this chapter, we discussed
the antiviral drug discovery for NiV. This chapter is majorly focused on two aspects,
one is experimental novel or repurposed drugs and the other is the computational
approaches to identify inhibitors for NiV.
5 Targeted Computational Approaches to Identify Potential Inhibitors … 139

5.2 Experimentally Tested Repurposed Drugs or Novel

Molecules Against NiV

Drug repurposing is an approach used for identifying different uses of already existing
licensed or investigational or approved medicines [13]. It provides several advantages
like low risk of failure, reduced time frame, and low investment to develop repurposed
drugs [14]. Therefore, several research groups made an attempt to discover potent
drugs either using repurposed drugs or develop novel molecules against NiV. For
example, Bitsch et al., in 2001, conducted an open trial in 194 patients; 140 out of
194 were treated with ribavirin, and 54 were treated as control. In the case of treated
patients, 128 patients were given oral ribavirin as 2 gm on day 1, 1.2 gm on day 2–4,
1.2gm on day 5–6, and 0.6gm on day 7–10, and 12 patients were given intravenous
ribavirin as 30 mg/kg, 16 mg/kg in every 6 h for 4 days, and then 8 mg/kg in every
8 h for 3 days. Forty-five deaths (~32%) were observed in the treated group, and 29
deaths (~54%) were observed in the control group. Some frequent side effects were
also observed, such as teratogenic effects, anemia, and jaundice. Overall, treatment
with ribavirin in patients with Nipah encephalitis reduced mortality by 36% [15].
Georges-Courbot et al., in 2006, tested the activity of ribavirin, 5-ethynyl analogue
(EICAR) of ribavirin, 6-aza-uridine, pyrazofurin, and poly(I)-poly(C12 U) against
NiV in vitro as well as in vivo. In in vitro testing, effective concentration of ribavirin,
5-ethynyl analogue (EICAR) of ribavirin, 6-aza-uridine, pyrazofurin, and poly(I)-
poly(C12 U) were found to be 100 μg/ml, 1 μg/ml, 0.25 μg/ml, 0.12 μg/ml, and
6.25 μg/ml, respectively. Then, they performed an in vivo toxicity assay to test the
toxicity of EICAR and 6-aza-uridine in hamsters. In the case of EICAR, all animals
were found to be healthy with normal body weight gain, whereas, in the case of
6-aza-uridine, all animals lost weight and were found dead within 6 days. Due to
the less availability of some of these compounds, further experiments could not be
performed. In the first experiment of efficacy testing, they gave ribavirin dose of
50 mg/kg/day to six animals, 6-aza-uridine dose of 175 mg/kg/day to six animals,
and phosphate-buffered saline as controls to six animals. The mean day of death was
found to be 5.1 ± 0.7 days for controls, 6.1 ± 0.7 days for 6-aza-uridine, and 6.8 ±
0.7 days for ribavirin. As no therapeutic effect was observed for the ribavirin, they
decided to use a lower concentration of ribavirin in the second experiment. They
gave ribavirin dose of 50 mg/kg twice daily to six animals, poly(I)-poly(C12 U) dose
of 3 mg/kg once daily to six animals, and placebo to six animals. The mean day of
death was found to be 7.3 ± 2.94 days for controls, 8.6 ± 2.94 days for ribavirin, and
all animals were alive till the 14th day for poly(I)-poly(C12 U). Seven samples (two
controls, two ribavirin-treated animals, and three 6-aza-uridine treated animals) were
collected after death for checking the presence of NiV. Seven samples (two controls,
three ribavirin-treated animals, and two 6-aza-uridine treated animals) were collected
within the initial 6 days of infection and examined for the presence of NiV. Only one
sample of 6-aza-uridine treated was found positive for anti-NiV IgG in the samples
collected within the initial 6 days of infection. Moreover, they checked the lungs,
livers, spleens, kidneys, and brains of all dead animals for the presence of NiV by
140 S. Gautam and M. Kumar

real-time RT-PCR and by measuring infectious viral titers. The virus was found to
be present in all the dead animals by the real-time RT- PCR, and the viral titer was
not detectable in three samples only. From the second experiment, one sample was
collected from the ribavirin-treated animals, and six samples were collected from
the poly(I)-poly(C12 U)-treated animals. One poly(I)-poly(C12 U)-treated animal was
sick for some reason and therefore not euthanatized, and the other five poly(I)-
poly(C12 U)-treated animals were euthanatized on the 30th day after infection. The
virus was not detected in any of the five poly(I)-poly(C12 U)-treated animals, whereas
viral RNA was found in the liver of three animals and the kidney of one animal [16].
Aljofan et al., in 2009, developed an assay of high throughput screening (HTS)
for the identification of antivirals against live NiV and HeV. They single blindly
screened a library containing 8040 low molecular weight compounds. For the
screening purpose, they used 10 μM compounds in triplicates and 90% efficacy
of compounds as the minimum criteria for lead compounds. They identified 28
molecules as lead compounds and further characterized their IC50 and CC50 values.
Three compounds out of 28 lead compounds were pointed out as gentian violet (also
called crystal violet), brilliant green (also called malachite green), and gliotoxin. All
three compounds effectively blocked the NiV and HeV. IC50 values of gliotoxin and
brilliant green were tenfolds lower than ribavirin, whereas IC50 values of gentian
violet were fourfolds lower than ribavirin for NiV. IC50 values of gliotoxin and bril-
liant green were found to be threefolds lower than ribavirin. In contrast, gentian
violet was found to be a bit less effective as compared to ribavirin for HeV. All three
compounds were more cytotoxic than ribavirin in Vero cells. All three compounds
were found to be more effective against NiV than HeV based on the therapeutic index
(TI) value. Antivirals’ potency was also checked against parent pseudotyped virus
(pVSV), influenza H1N1 virus, human parainfluenza virus type 3(HPIV3), pseudo-
typed Nipah virus (pNiV), and pseudotyped Hendra virus (pHeV). Similar inhibition
rates were observed for these viruses except for gliotoxin inhibition for influenza virus
in a dose-dependent manner. Further, they performed the real-time PCR to check the
response of these compounds on the inflammatory response profile, i.e., changes in
the cytokines IL-8 and TNF-α levels. They observed that brilliant green increases
the IL-8 and TNF-α levels by 15–20 folds. Gliotoxin and gentian violet induced
IL-8 levels by twofold and decreased TNF-α levels. All three compounds are used
in antibacterial and antifungal therapy. They are unsuitable for internal intake; there-
fore, they can be used as disinfectants, positive control for experimental studies, and
topical antiviral applications [17].
Pallister et al. in 2009 evaluated chloroquine’s effectiveness, a known antimalarial
agent in preventing lethal NiV infection in vivo using ferrets. Earlier research showed
that chloroquine could prevent NiV infection at a concentration of ≥1 μM in vitro.
For the in vivo study, they give 25 mg/kg/day dose of chloroquine to three ferrets
24 h before viral exposure and give 25 mg/kg/day dose of chloroquine to three
ferrets 10 h after viral exposure. One control ferret was given 20% sucrose 24 h
before viral exposure, and one control ferret was given 20% sucrose 10 h after viral
exposure. The virus isolate used in this study was NiV-Malaysia isolate EUKK 19817.
They observed no change in the pathology, virus distribution, viral RNA levels, and
5 Targeted Computational Approaches to Identify Potential Inhibitors … 141

replication process of the control and the treated animals. Chloroquine concentrations
in serum were higher than their requirement for anti-NiV activity in vitro. On the
whole, chloroquine was not an effective drug for treating NiV infection [18].
Freiberg et al., in 2010, tested the combined treatment of chloroquine and ribavirin
in the golden hamster model infected with NiV and HeV. They first checked in vitro
activity of chloroquine and ribavirin in HeLa cell lines. Effect of drug treatment was
checked by measuring viral titers in the supernatants of cells infected with the virus.
When chloroquine was given, IC50 was found to be 0.71 and 0.62 μM for HeV and
NiV, and in the case of ribavirin, IC50 was found to be 4.96 and 4.18 μM for HeV
and NiV. Then, they checked the in vivo activity of chloroquine and ribavirin in the
golden hamster model infected with NiV and HeV. For this, five animals were given
ribavirin (30 mg kg−1 in every 12 h), five animals were given chloroquine (50 mg kg−1
every day), and five animals were given a combination of both chloroquine and
ribavirin. One control group is virus-infected animals having vehicle solution only,
and the other control group has drugs only. Ribavirin-treated NiV-infected hamsters
delay death by up to 5 days but cannot prevent death, whereas no positive impact
was seen in ribavirin-treated HeV-infected hamsters. Chloroquine-treated NiV and
HeV-infected hamsters neither delay nor prevent the death, whereas it reduces the
survival by 3 days for NiV and 2 days for HeV. Chloroquine in combination with
ribavirin cannot increase the mean survival times of the animals compared to the
control group. There is no clear-cut effect observed in either case of treatment, i.e.,
individually or in combination, so they checked whether high doses of these drugs
would be able to provide positive outcomes or not. They provided ribavirin and
chloroquine in three different doses, i.e., 100, 150, and 200 mg kg−1 day−1 and 50,
100, or 150 mg kg−1 day−1 . In the case of ribavirin, 100 mg kg−1 day−1 was the
most effective concentration for survival and delayed death by 3 days after infection.
About 150 and 200 mg kg−1 day−1 were less effective as these concentrations only
delayed death by 1 day. The maximum tolerable dose of ribavirin was between 150
and 200 mg kg−1 day−1 . In the case of chloroquine, 50 mg kg−1 day−1 was found
to be the most effective concentration for survival and delayed death by 5 days
after infection. About 100 and 150 mg kg−1 day−1 were found to be less tolerated
and lead to death by day 2. Altogether, chloroquine and ribavirin individually or in
combination could not protect virus-infected hamsters [19].
Mohr et al., in 2015, examined the 163 kinase inhibitors targeting the PI3K/Akt
pathway, hypothesizing that cellular kinases are necessary for the replication process
in various BSL-4 viruses. They initially tested all these inhibitors to check their ability
to block Lassa virus (LASV). They found out that eight compounds including control
exhibited anti-LASV activity. Then, these eight compounds were checked for their
ability to inhibit replication against Marburg virus (MARV-GFP), Alkhurma hemor-
rhagic fever virus (AHFV), Nipah virus (NiV-luc recombinant reporter viruses),
Ebola virus (EBOV-GFP), and Crimean Congo hemorrhagic fever virus (CCHFV).
Two compounds-OSU-03012 (also called AR-12) and BIBX 1382 dihydrochloride
out of eight compounds were effective against LASV as well as other viruses. OSU-
03012 inhibited replication of LASV with an EC50 of 0.5 μM; CC50 in the range of
3.1–5.7 μM and SI lied between 6 and 11, respectively. It also inhibited EBOV-GFP,
142 S. Gautam and M. Kumar

NiV-luc, and MARV-GFP with an EC50 of 0.3, 0.4, and 0.3 μM, CC50 of 6.4, 8.2,
and 7.1 μM, and SI of 23, 21, and 20, correspondingly. No inhibition effect of OSU-
03012 was observed on AHFV and CCHFV. BIBX 1382 dihydrochloride inhibited
replication of MARV-GFP, LASV, and EBOV-GFP with an EC50 of 1.8, 3.2, and
1.1 μM, CC50 of 29.1, 15.3, and 19.8 μM, and SI of 19, 6, and 18. No inhibition
effect of BIBX 1382 dihydrochloride was observed on CCHFV, NiV-Luc, and AHFV.
Further, they examined the effect of OSU-03012 and BIBX 1382 dihydrochloride on
viral titers of wild-type EBOV and LASV. They observed that both were reducing
wild-type EBOV and LASV titers by 2–3 logs in a dose-dependent manner. Then,
to validate the antiviral effect of OSU-03012 and BIBX 1382, they measured the
RNA of LASV in various cell lines using qRT-PCR. Both OSU-03012 and BIBX
1382 decreased RNA levels of LASV in a dose-dependent manner in all the cell lines.
Moreover, they also studied the mechanism of action of OSU-03012 and BIBX 1382.
In this, firstly they checked whether these compounds are inhibiting the entry step of
EBOV and LASV with the help of HIV particles pseudotyped with the viral glyco-
proteins. OSU-03012 was not able to block the glycoprotein-mediated entry step of
both the viruses, whereas BIBX 1382 was able to block the glycoprotein-mediated
entry step of both the viruses. Secondly, they checked that OSU-03012 and BIBX
1382 affect the formation of virus-like particles (VLP) of LASV Z protein. Both
OSU-03012 and BIBX did not inhibit the release of VLP of LASV Z protein [20].
Rashighi and Harris in 2017 checked the antiviral potency of 4' -Azidocytidine
(R1479), known to inhibit dengue virus, hepatitis C virus (HCV), and respiratory
syncytial virus (RSV) by blocking the RNA-dependent RNA polymerase (RdRP)
function. Because of the RdRP binding domain structure conservation all-over
multiple families of viruses, they explored the antiviral activity against NiV and
HeV. Firstly, they checked the potential of R1479 to block the reporter activity of
luciferase or recombinant GFP expressing NiVs. From this, EC50 values for NiV-
Luc2AM and NiV-GFP2AM were found to be 1.12 and 1.64 μM. Then, they checked
the potential of R1479 to block the cytopathic effect (CPE) induced by wild-type
NiV and HeV infections in NCI-H358 cells at 72 h post-infection (hpi) using a cell
viability reduction assay. From this, EC50 values for NiV and HeV were found to
be 2.63 and 1.75 μM. They also checked the potential of R1479 to block other
viruses of the Paramyxoviridae family such as measles virus, human parainfluenza
virus, mumps virus. Secondly, they assessed the inhibition ability of R1479 to block
henipaviruses using immunofluorescence assays. Lastly, they performed a virus titer
reduction (VTR) assay to check the potential of R1479 to block NiV and HeV and
found EC50 values of 1.53 and 2.41 μM, respectively. These multiple assays showed
that R1479 could efficiently inhibit henipaviruses with lower EC50 values (μM) [21].
Dawes et al., in 2018, evaluated the potential of favipiravir against NiV and HeV
in vitro and assessed its ability in vivo against NiV-M. Favipiravir (T-705, 6-fluoro-3-
hydroxy-2-pyrazinecarboxamine) is a purine analogue, RdRp inhibitor, and is already
approved for use against influenza strains in Japan. It also exhibited antiviral activity
against various viruses such as bunyaviruses, filoviruses, flaviviruses, enteroviruses,
arenaviruses, noroviruses, alphaviruses, and rhabdoviruses. It has been checked
in vitro for paramyxoviruses such as measles virus, human parainfluenza virus
5 Targeted Computational Approaches to Identify Potential Inhibitors … 143

3, avian metapneumovirus, respiratory syncytial virus, human metapneumovirus

(hMPV), and Newcastle disease virus and in vivo against human metapneumovirus
(hMPV). In this study, they checked the ability of favipiravir to block the NiV-
Bangladesh (NiV-B), HeV, NiV-Malaysia (NiV-M), recombinant NiV expressing
Gaussia luciferase, and eGFP (rNiV-Gluc-eGFP) using virus yield reduction assay
in vitro. From this assay, EC50 values for NiV-B, HeV, NiV-M, and rNiV-Gluc-eGFP
were found to be 14.82, 11.71, 44.24, and 14.57 μM, respectively. Selectivity index
(SI) for NiV-B, HeV, NiV-M, and rNiV-Gluc-eGFP was found to be >67.47, >85.39,
> 22.60, and >66.63. EC90 values for NiV-B, HeV, NiV-M, and rNiV-Gluc-eGFP
were found to be 15.87, 16.49, 123.8, and 16.25 μM. Then, they assessed the effect
of delayed favipiravir treatment on NiV infection by measuring luciferase activity at
different time periods. They observed that favipiravir is able to inhibit NiV infection
effectively when it is added instantly. Further, they performed in vivo studies by giving
favipiravir orally and subcutaneously in the Syrian hamster model. When favipiravir
was given orally instantly after infection twice daily for 14 days, they observed high
levels of viral P gene expression in controls compared to treated animals. Neutralizing
antibody levels were measured in terms of PRNT50 , which were >80 and >1280 in two
treated animals and <20 levels in three treated animals. These observations showed
that orally favipiravir is highly effective in preventing NiV when given instantly after
infection. When favipiravir was given subcutaneously once daily for 14 days, they
observed similar results. Lastly, they evaluated the pathological changes associated
with the favipiravir-treated NiV-M-infected hamster model. For this, they inspected
the lung, spleen, and brain of animals taken during the study and the survivors at
42 days post-infection using H&E staining and IHC techniques targeting NiV nucle-
oproteins. Vehicle-treated animals showed various symptoms such as perivascular
infiltration of inflammatory cells, presence of NiV antigens in endothelial cells,
interstitial pneumonia with alveolar edema or hemorrhage in lungs, red pulp cord
displayed necrotic areas scattered with mononuclear or reticular cells with NiV anti-
gens in spleen and meningitis with infiltration of neutrophils and mononuclear cells
in the brain. Favipiravir-treated animals did not show any critical disease-associated
findings, and no NiV antigens were present in the lung, spleen, and brain of animals
[22].
Lo et al., in 2019, checked the efficacy of remdesivir against NiV-Bangladesh
(NiV-B) genotype in African green monkeys (AGM). Remdesivir is a nucleotide
analog prodrug exhibiting broad-spectrum antiviral activity against filoviruses,
paramyxoviruses, and coronavirus. In this study, the first groups with four AGMs
(two males and two females) were infected with a lethal dose of NiV-B through
the intratracheally and intranasally route and, after 24 h, treated with remdesivir
once daily for 12 days. The second groups with four AGMs (two males and two
females) were infected with a lethal dose of NiV-B through intratracheally as well as
intranasally route and, after 24 h, treated with a vehicle only once daily for 12 days.
Vehicle-treated animals initially developed mild respiratory symptoms and progress
toward respiratory distress, increased respiration rate, and decreased oxygen satu-
ration but no change in body weight and temperature. Remdesivir-treated animals
showed reduced appetites and developed respiratory problems, but no change in
144 S. Gautam and M. Kumar

body weight, temperature, respiration rate, and oxygen saturation levels. Only one
remdesivir-treated animal was found to be positive for viral RNA by quantitative
reverse transcription-polymerase chain reaction (qRT-PCR); all others were nega-
tive during the experiment. The virus was found in the blood of all vehicle-treated
animals. For virological analysis, 35 tissues were taken from every animal, and high
viral loads were observed for vehicle-treated animals on the 7th and 8th day post-
infection (dpi). Viral RNA was found in 11 tissues taken from three remdesivir-treated
animals and in the brain of one remdesivir-treated animal at 92dpi. Further, histolog-
ical examination showed the slightest indication of focal perivascular inflammation
and absence of NiV antigen checked by immunohistochemistry (IHC). The animal
in which viral RNA is present in the brain showed mild-to-moderate mononuclear
meningoencephalitis. Lately, they checked the virus-neutralizing titers in the serum
of remdesivir-treated animals collected at different time points. They observed low
virus-neutralizing titers in all the animals by 19 dpi [23].
Lo et al., in 2020, checked the in vitro activity of β-D-4' -chloromethyl-2' -deoxy-
'
2 -fluorocytidine, known to treat pediatric and adult respiratory syncytial virus infec-
tion and are currently under phase I and phase II clinical trials. β-D-4' -chloromethyl-
2' -deoxy-2' -fluorocytidine (ALS-8112) is a parent nucleoside of the methyl propi-
onate derivative lumicitabine (ALS-008176, JNJ-64041575). Their blind screening
against recombinant reporter NiV expressing ZsGreen1 fluorescent protein (rNiV-
ZsG) led them to find ALS-8112 as a hit compound for testing. Firstly, they checked
the ability of ALS-8112 to inhibit rNiV-ZsG based on a green fluorescence signal,
and they observed that ALS-8112 was able to block rNiV-ZsG in a dose-dependent
manner with CC50 of >100 μM. They also checked the potential of ALS-8112 to
inhibit other human respiratory viruses such as GFP expressing recombinant RSV
(rgRSV224), measles virus (rMVEZ-GFP), human parainfluenza virus 3 (hPIV3-
GFP), and rNiV-ZsG. They used two human respiratory epithelial cell lines—NCI-
H358 and HSAEC1-KT. They checked the ability of ALS-8112 to inhibit viruses
based on green fluorescence signal and found that it inhibited rgRSV224 in both
the cell lines with EC50 of 1.23 μM in NCI-H358 and 0.36 μM in HSAEC1-KT
cells. It also inhibits rNiV-ZsG with EC50 of 0.56 μM in NCI-H358 and 0.84 μM
in HSAEC1-KT cells, whereas very low potent for hPIV3-GFP and rMVEZ-GFP(3)
in both the cell lines. Secondly, as it is known that NiV produces in vitro cytopathic
effect (CPE), so they checked the ability of ALS-8112 to block the CPE produced
by wild-type NiV-Malaysia (NiV-M) genotype, NiV-Bangladesh (NiV-B) genotype,
and rNiV-ZsG and measured by the cellular ATP levels luminescence. They found
that it inhibited NiV-produced CPE in both the cell lines with EC50 in the range
of 0.89–3.08 μM, CC50 of above 50 μM, and also calculated SI using CC50 and
EC50 values. Lastly, they checked the ability of ALS-8112 to reduce the infectious
virus titers against NiV-B and rNiV-ZsG. From this, they observed a 6 or 7 times
reduction in virus titers of NiV-B and rNiV-ZsG, respectively. Moreover, they also
captured fluorescence micrographs of infected cells with rNiV-ZsG with different
concentrations of ALS-8112 at 48 hpi. In this, they observed that infected cells
were decreasing in a dose-dependent manner and completely ablated infected cells
at the concentration of 12.5 μM. Similarly, they checked the ALS-8112 toxicity
5 Targeted Computational Approaches to Identify Potential Inhibitors … 145

in different cell lines such as primary human peripheral blood mononuclear (PBM),
human epithelial lung (A549), human lymphoblastoid (CEM), human hepatocellular
carcinoma cells (HepG2), and Vero cells. There was no toxicity observed in HepG2,
Vero cells, and A549 cells, whereas toxicity is observed in PBM and CEM cells
with CC50 of 4.2 μM and 2.8 μM, respectively. In vivo testing of nucleoside analogs
generally results in various side effects such as pancreatitis, anemia, lactic acidosis,
neutropenia. Due to these side effects, they measured the mitochondrial toxicity,
bone marrow toxicity, and lactic acid production, as these are stress markers of the
cells [24].
Several groups also used other methods to find effective therapy such as mono-
clonal antibodies, peptides, conjugated peptides, RNA interference (RNAi). Like
Zhu et al., in 2006, reported the potent neutralizing humanized monoclonal anti-
bodies (hMAbs) against the viral envelope glycoprotein (G) of NiV and HeV. They
used soluble, purified, and oligomeric HeV G as the antigen for screening an exten-
sive naive phage display library to identify potent antibodies. After multiple rounds
of the panning process, seven hMAbs were selected based on the binding activity
to a soluble HeV G. They performed a cell fusion assay to assess the potential of
these antibodies in inhibiting entry and membrane fusion. Fab m101 was found to be
the most potent cell fusion inhibitory activity, whereas m102 and m106 were found
to display cross-reactivity. Conversion of m101 to Immunoglobulin G1 (IgG1) was
responsible for exceptionally high cell fusion inhibition activity. About 12.5 μg/ml
of IgG1 m101 was required to neutralize 100% infectious HeV, and only 1.6 μg/
ml of IgG1 m101 was required to neutralize 98% infectious HeV. m101, m102,
and m103 antibodies were competing, thus suggesting that these antibodies exhibit
overlapping epitopes. This study showed that these humanized antibodies are new
immunotherapy for treating HeV and NiV [25].
Similarly, Dang et al., in 2019, reported the antibody that targets the fusion
glycoprotein (F) to inhibit the NiV and HeV infection. They performed the
cloning, sequencing, and generated humanized antibody of 5B3 (h5B3.1). They also
performed a neutralization assay showing that 5B3 and h5B3.1 potently inhibited
NiV and HeV infection. Further, they determined the structure of the 5B3 anti-
body complex with NiV-F trimer using cryogenic-electron microscopy (cryo-EM).
Complex structural analysis showed that 5B3 antibodies recognize perfusion-specific
different epitopes, which are conserved in both NiV and HeV F. Overall, 5B3 antibody
could be an effective therapy against Henipaviruses (HNVs) [26].
Likewise, Dang et al., in 2021, reported the neutralizing mouse monoclonal anti-
bodies, i.e., 12B2 and 1F5, against the NiV and HeV fusion glycoprotein (F). They
showed that both 12B2 and 1F5 antibodies bind the F protein with strong affinity
and neutralize NiV and HeV in the BSL-4 containment. They determined the struc-
ture of the 12B2 antibody complex with NiV-F and 1F5 antibody complex with
HeV F at 2.9 and 2.8 Å resolutions using cryogenic-electron microscopy (cryo-EM).
Complex structural analysis showed that both the antibodies recognize perfusion-
specific different epitopes, which are conserved in both NiV and HeV F. Further,
they performed membrane fusion assay, which showed that both 12B2 and 1F5
antibodies hold the F protein in the perfusion conformation and block the changes
146 S. Gautam and M. Kumar

that are essential for the membrane fusion. They also generated humanized 12B2
(h12B2) and 1F5 (h1F5) antibodies and assessed their neutralization ability against
NiV-Malaysia (NiV-M), NiV-Bangladesh (NiV-B), and HeV. For this, they calcu-
lated IC50 using plaque reduction assay to determine virus neutralization with the
increasing concentration of antibodies. They found that IC50 is in the range of 0.4–
3.6 μg/ml for 12B2–h12B2 and 0.2–1.3 μg/ml for 1F5–h1F5 against NiV-B, NiV-M,
and HeV. This study showed that antibodies could be an effective therapy against
henipaviruses (HNVs) [27].
Mathieu et al., in 2018, engineered the new antiviral lipopeptides and assessed
their efficacy in vitro and in vivo against NiV. They used the known “VIKI” sequence
to engineer new antiviral lipopeptides and developed various lipopeptides such as
VIKI-dPEG4-Toco, VIKI-dPEG4-Chol, VIKI-dPEG4, VIKI-dPEG4-bisToco, and
VIKI-dPEG4-bisChol. Then, they checked these lipopeptides’ protease sensitivity
and observed that the presence of cholesterol or tocopherol increases the resistance to
degradation by proteases. Based on this observation, they used VIKI-dPEG4-Toco
and VIKI-dPEG4-Chol for further experiments. They correlated the effectiveness
of VIKI-dPEG4-Toco and VIKI-dPEG4-Chol in hamsters against NiV. For this,
they infect the hamsters with 106 pfu NiV and then treat intranasally with 10 mg/
kg peptide or vehicle on days—1, 0, and 1 of infection. They observed that treat-
ment with peptides improves survival and 100% death in untreated animals. Based
on biodistribution, VIKI-dPEG4-Toco was further used to check the effectiveness
of VIKI-dPEG4-Toco in African green monkeys (AGMs). For this, they infect the
hamsters intratracheally with 2 × 107 pfu of NiV, then treated with 10 mg/kg peptide
intratracheally and 2 mg/kg peptide subcutaneously on days—1, 0, and daily for
5 days. They observed that treatment with peptides leads to the protection from
lethal outcomes. This study showed that conjugated peptides could effectively treat
lethal NiV infection [28].
Mungall et al., in 2008, designed eight (four large polymerase (L) and four nucleo-
capsid (N) gene-specific) siRNA molecules against NiV, two N gene-specific against
HeV, and two siRNA molecules as a control. They tested the ability of these siRNA
molecules to block a henipavirus minigenome replication system and live virus
in vitro. Three out of four L gene-specific siRNA could inhibit replication using
the minigenome system. N gene-specific siRNA inhibits only live virus replica-
tion, indicating that targeting early expressed gene transcripts is more effective than
late expressed gene transcripts. siRNA molecules targeting NiV infection were only
partially effective in inhibiting HeV infection. Overall, this study illustrated that
inhibiting henipavirus by RNA interference approach could be an effective therapy
[29].
5 Targeted Computational Approaches to Identify Potential Inhibitors … 147

Fig. 5.1 Computational/in

silico-based approaches to
identify the potent drugs
against NiV

5.3 Computational Approaches for the Identification

of Antiviral Drugs for NiV

Many research groups made an effort to find the potential drugs using novel and
repurposed drugs but still lack licensed drugs or vaccines, so there is a need to speed up
the drug discovery process using various computational approaches. Several groups
have used approaches as given in Fig. 5.1 to find the effective antivirals against NiV.

5.4 Machine Learning and QSAR-Based Prediction

Approach

Machine learning is one of the important approaches in identifying inhibitors against

various viruses. Several machine learning-based antiviral predictors using quan-
titative structure activity relationship (QSAR) information of molecules/peptides
are available such as Anti-Ebola [30], anticorona [31], HIVprotI [32], anti-flavi
[33], AVCpred [34], AVP-IC50 Pred [35], AVPpred [36]. Further, there are various
antiviral databases which are available such as DrugRepV [37], AVPdb [38]. For
NiV, Rajput et al., in 2019, developed a QSAR-based predictor and integrated it into
a user-friendly web server “Anti-Nipah”. Three hundred and thirteen experimentally
tested chemicals were extracted from literature, and finally, 95 non-redundant chem-
icals with their IC50 values were converted into pIC50 . Simplified molecular-input
line-entry systems (SMILES) of these chemicals were converted into a 3D-standard
data format (3D-SDF). Further 3D-SDF was used to calculate 17,967 descriptors
with the help of PaDel software. Then, these descriptors were used to fetch the most
essential 42 features using “RemoveUseless” followed by “CfsSubsetEval”. The
148 S. Gautam and M. Kumar

model was developed using a support vector machine (SVM) through a tenfold cross-
validation technique using these features. The model performance was evaluated
using Pearson’s correlation coefficient (PCC), Root mean absolute error (RMSE),
Coefficient of determination (R2), and Mean absolute error (MAE). Training/testing
and independent validation dataset showed the PCC of 0.82 and 0.92 during tenfold
cross-validation. The applicability domain analysis by William’s plot and scatter
plot checked the robustness of the developed model. William’s plot showed that all
the points of training/testing and independent validation dataset lie in the range of
threshold values of standard residues and leverage. A scatter plot between actual
pIC50 and predicted pIC50 showed that all the points of training/testing and indepen-
dent validation dataset lie near the trendline. Decoy set and chemical clustering using
RApid DEcoy Retriever (RADER) software and compounds-specific bioactivity
dendrogram (C-SPADE), respectively, also validated models robustness. Chemical
clustering analysis displayed that these compounds are diverse in nature. Highly
effective compounds with low IC50 formed clusters together and vice-versa. Some
highly and less effective compounds were also clustered together. “Anti-Nipah” web
server can be used for extracting information like inhibitors of the NiV given in
the literature and patents, predicting the antiviral activity of a query molecule, and
drawing the structure of query molecules [39].

5.5 Molecular Docking

Several research groups used molecular docking method to find the potent molecules
against NiV. Molecular docking is a kind of modeling approach to understand the
ligand and its target interactions [40]. Using this approach, Lipin et al., in 2021, exam-
ined the structure–property relationship of favipiravir, which is known to exhibit
exemplary in vitro activity against NiV, and designed a series of 15 piperazine-
substituted favipiravir derivatives and then computationally screened their interac-
tion and ability to inhibit NiV-G protein. Density functional theory analysis was
done to calculate all the derivatives’ geometrical features and electronic proper-
ties. ADMET and toxicity analysis were done using the SWISSADME server and
ProTox-II online tool, respectively. All the derivatives satisfied the Lipinski rule
of five, high gastrointestinal absorption (GI), good solubility, and non-toxic, thus
showing good oral bioavailability. PIC50 values were predicted using a web server,
which showed that predicted PIC50 values for favipiravir derivatives were better than
the parent favipiravir. Further molecular docking analysis and understanding of the
binding mode and the affinity of favipiravir derivatives toward the NiV-G (PDB
ID-3D11) were done using Maestro glide docking program and GLIDE—6.6. This
study showed that piperazine-substituted favipiravir derivatives could be promising
inhibitors against NiV [41].
Similarly, James et al., in 2021, prepared 22 derivatives of favipiravir having
pyrazine and other heterocyclic groups as moiety Nipah glycoprotein, i.e., 3D11 was
taken from protein data bank (PDB) for novel inhibitors against NiV using various
5 Targeted Computational Approaches to Identify Potential Inhibitors … 149

computational approaches. Molecular docking studies were performed using all 22

derivatives and Nipah glycoprotein, i.e., 3D11, to determine the various conforma-
tions of these complexes. Further analysis showed that 13 derivatives have higher
docking scores than the standard favipiravir; thus, it is proposed that these derivatives
might have a perfect affinity for the NiV proteins. Docking scores for compound
5_Favipiravir, 4_Favipiravir, and 19_Favipiravir were found to be −6.16, −5.50,
and −5.38 kcal/mol, respectively. These three compounds had pyrazole, imidazole,
and pyrazinone as heterocyclic groups. Further physicochemical properties’ analysis
showed that all derivatives’ properties lie in the expected value range, thus exhibiting
good oral bioavailability. Lastly, in silico ADMET studies showed that derivatives
have good scores for human oral absorption, human serum albumin binding, Caco-2
permeability, total solvent accessible surface area, etc. [42].

5.6 Molecular Dynamics

Some groups used molecular dynamics approach like Sen et al., in 2019, who used
various in silico approaches like homology modeling or ab initio modeling, peptide
designing, and molecular docking. Out of nine NiV proteins, four proteins, i.e., F, N,
G, and P partial structures, were taken from the protein data bank (PDB), and using
these structures, models were developed for the five proteins, i.e., M, W, V, L and F
proteins with the help of homology modeling, ab initio modeling, and threading. They
also contribute about 90% to the structural characterization of NiV proteins compared
to the structural data available in the PDB of NiV proteins. With the help of these
models, they designed four potent peptide inhibitors (one against F protein trimer,
one against M protein dimer, and two against G-Protein-human ephrin-B2 receptor).
Three independent molecular dynamics simulations were performed of 100 ns to
check the stability of these four protein–peptide complexes. They also screened
22,685 compounds of the ZINC library against NiV proteins using AutoDock4 and
Dock6.8 programs. Finally, predicted 146 small molecules as inhibitors that can bind
G, F, P, N, and M NiV proteins and then cut down to 13 molecules. Three indepen-
dent molecular dynamics simulations were performed of 50 ns to check the stability
of these 13 protein–peptide complexes. They also determined the binding energies
of these complexes with the help of MM/PBSA and found out that nine complexes
had negative energy, one had positive energy, and three could not bind the proteins.
Molecular docking studies also showed that few proposed inhibitors are already
tested as repurposed drugs. For example—ZINC04829362 (Cyclopent-1-ene-1,2-
dicarboxylic acid) is already known as an antiasthmatic and antipsoriatic drug and
ZINC12362922 (Bicyclo[2.2.1]hepta-2,5-diene-2,3-dicarboxylic acid) known drug
for depression and Parkinson’s disease. Furthermore, they checked the effectiveness
of recommended inhibitors against 15 (seven Malaysian, three Bangladeshi, and five
Indian) strains of the NiV available in the NCBI Database. They checked the varia-
tions with the help of Multiple Sequence Alignment using MUSCLE and reduced it
to only those variations that were in immediate contact with the inhibitors. Out of five
150 S. Gautam and M. Kumar

residues changes (Lys236Arg, Asp188Glu, Gln211Arg, Asp252Gly, and Ile331Val),

four were conservative substitutions, and one (Asp252Gly) was found to be a non-
conservative change. Results showed that these recommended inhibitors could be
potential antivirals against every single NiV strain [43].
Ropón-Palacios et al., in 2020, used virtual screening techniques like molecular
docking and molecular dynamics to identify the potential novel antivirals against
the NiV. One hundred and eighty-three ligands were taken from “The Pathogen Box
Medicines for Malaria Venture (MMV, Geneva, Switzerland)” and allowed them to
interact with NiV glycoprotein (NiV-G). NiV-G is involved in the entry of the virus by
binding to the Ephirin-B2 (EFNB2) and Ephirin-B3 (EFNB3) receptors present on the
surface of the host cells, therefore a promising target for inhibiting the virus infection.
Of 183 ligands, three (MMV020537, MMV019838, and MMV688888) were potent
inhibitors with binding energies of −11.8, −9.5, and −9.2 kcal/mol using a virtual
screening approach. To refine the results of virtual screening, molecular docking
was carried out with the Lamarkian hybrid genetic algorithm available in AutoDock.
MMV020537 showed the binding energy of −14.29 kcal/mol (Kd = 0.03 nM),
MMV019838 showed the binding energy of −10.23 kcal/mol (Kd = 31.61 nM),
and MMV688888 showed the binding energy −11.82 kcal/mol (Kd = 2.18 nM). In
both virtual screening and molecular docking, Ligand 1, i.e., MMV020537, had the
lowest interaction energy. They also identified the two new residues, i.e., Cys240
and Arg236, present in the binding site and involved in the ligand recognition at
3.1 and 1.9 Å. Validation of molecular docking was done using X-Score as well as
PLANTS software. Further molecular dynamics simulation studies showed that a
complex formed between first ligand and NiV-G protein is found to be stable during
production time (40 ns) [44].
Kalbhor et al., in 2021, took three chemical library databases—Asinex-Antiviral
Library (8722), Enamine-Antiviral Library (3700), and ChemDiV-Antiviral Library
(67,470) containing a total of 79,892 chemicals and NiV-G protein complex bound
with cell surface receptor ephrin-B2 (PDB ID-2VSM) available in the Protein data
bank (PDB) which were taken for analysis. Multi-step molecular docking analysis
was performed to get those compounds docked with NiV-G protein using Glide-
HTVS, Glide-SP, and Glide-XP. For further analysis, 299 compounds were selected
based on the XP dock score and MM-GBSA score. Further pharmacokinetic analysis
like ADME and synthetic accessibility properties have been carried out using 299
compounds. About 207 out of 299 compounds were found to be good with drug-
likeness parameters like absorption, distribution, metabolism, and excretion. Then,
toxicity-based analysis was performed using TOPKAT and found that 14 compounds
are non-toxic in nature. Moreover, their molecular binding modes and intermolecular
interactions were checked to cut down the compounds, and the final five compounds
were obtained as NiV-G protein modulators. Then, they deeply explored the molec-
ular binding interaction analysis of NiV-G protein and proposed inhibitors using
the XP-docking method and protein–ligand interaction profiler (PLIP) tool. Molec-
ular binding interaction analysis showed that Tyr581 residue of NiV-G protein was
found to be a common residue involved in the H-bonding with various compounds,
and Lys560, Gln559, Tyr581, Val507, Glu579, and Ile588 residues of NiV-G protein
5 Targeted Computational Approaches to Identify Potential Inhibitors … 151

were found to be common residues involved in the hydrophobic interactions with G1,
G2, G3, and G5 compounds. Moreover, they checked the ADME parameters, which
showed that all five compounds exhibited drug-like properties like molecular weight
of less than 500 g/mol, moderate-to-high soluble nature, orally active, good synthetic
accessibility score, etc. Toxicity profiling using TOPKAT tool suggested that all five
compounds are non-carcinogenic, non-toxic, and non-mutagenic. They also built
an EGG-BOILED model to analyze two more essential parameters, HIA (Human
Intestinal Absorption) and BBB (Blood–Brain Barrier). Further, they performed an
MD simulation analysis of 100 ns to evaluate the stability, as well as the dynamic
behavior of these NiV-G protein complexes using various parameters like root mean
squared deviation (RMSD), radius of gyration (RoG), and root mean square fluc-
tuation (RMSF). They also determined Molecular Mechanics Poisson–Boltzmann
Surface Area (MM-PBSA) to determine binding free energies (∆G) from all the
MD simulations to deduce the energy contribution of recommended inhibitors in
stabilizing the NiV-G protein complexes. Potential proposed inhibitors exhibit high
negative ∆G values in the range of −166.246 to −226.652 kJ/mol and showed strong
affinity toward the NiV-G protein complex [45].
Ahmed Bhuiyan et al., in 2022, took 92 compounds from Ambinter, but two
compounds were duplicates, so the final 90 compounds and NiV-G protein (PDB
ID-2VSM) from the RCSB protein data bank were used in this study. They predicted
the active site of NiV-G with the help of the Computed Atlas for Surface Topog-
raphy of Proteins (CASTp) server. Then, they performed molecular docking anal-
ysis using AutoDock Vina in which the top five compounds (CID: 11,096,158, CID:
11,861,102, Amb35795905, CID: 102,601,745) were selected, including control
(CID: 24,139) based on binding affinity scores for further analysis. Pharmacokinetic
properties and toxicity profiling of all five compounds were determined utilizing
the SwissADME server and pkCSM server, which showed that all five compounds
are suitable and non-toxic. Further molecular dynamics (MD) studies showed that
all protein–ligand complexes are stable, but Amb33921182, i.e., 2- acetamido-2-
deoxy-D-gluco-hexopyranose compound is the most potent drug candidate. They
also calculated root mean square fluctuation (RMSF), root mean square deviation
(RMSD), ligand properties, and protein–ligand contacts (P-L contact) [46].
Vinay Randhawa et al. in 2022 performed multi-target molecule screening using
molecular docking and molecular dynamics approach to find the potential anti-NiV
drugs targeting NiV-F, NiV-N, and NiV-G proteins. Potent known NiV inhibitors such
as drugs, phytochemicals, and small molecules were extracted from the literature
search using PubMed. Three-dimensional structures of drugs and small molecules
were drawn manually using MarvinSketch v5.10.0 software (chemaxon.com), and
3D structures of phytochemicals were taken from Serpentina database. Three-
dimensional structures of target proteins, i.e., NiV-F (PDB ID-5EVM), NiV-G (PDB
ID-2VSM), and NiV-N (PDB ID-4CO6), were downloaded from RCSB protein
data bank (PDB). Molecular docking studies were performed using QuickVina
v 2.0 software, and then eight molecules (two chemicals and six phytochemi-
cals) were selected for further analyses based on binding energy threshold values.
ADME and pharmacokinetic properties were computed using ADMETlab web
152 S. Gautam and M. Kumar

server, and three molecules were selected, i.e., two phytochemicals’ molecules—
CARS0358 and RASE0125—and one chemical-ND_nw_193.2 based on the z-
score. Further molecular dynamics simulations were carried out with target only
and with protein–ligand complexes for 5 ns. As a whole, two phytochemicals, i.e.,
CARS0358 (NA) and RASE0125 (17-O-Acetyl-nortetraphyllicine) were found to
inhibit all the three targets of NiV, whereas one chemical-ND_nw_193 (RSV604) was
found to inhibit NiV-N and NiV-G. CARS0358 (NA) and RASE0125 (17-O-Acetyl-
nortetraphyllicine) are indole alkaloids that inhibit Zika and dengue virus infec-
tion. ND_nw_193 (RSV604) is a chemical drug that inhibits the human respiratory
syncytial virus (RSV) [47]

5.7 Integrated Structure- and Network-Based Approach

Few researchers used integrated structure- and network-based approach like Pathania
et al., in 2020, who used an integrated structure- and network-based drug discovery
approach to identify the potential entry inhibitors for the NiV. For molecular docking,
NiV-G crystal structure (2VSM) was downloaded from the protein data bank (PDB),
and a small molecule library was prepared using 2327 Food and Drug Administration-
approved drugs (FDA-approved drugs) taken from the DrugBank database. Then,
structural optimization was performed with the help of the MMFF94 force field
available in the OpenBabel v 2.4.0 program. Moreover, applying reasonable charges
and hydrogen was added using AutoDock tools. For validating the molecular docking
approach, CCDC/ASTEX datasets containing 305 protein–chemical complexes were
refined to 265 complexes. Docking simulations utilized rigid receptors, flexible
ligands, and a grid box around the binding pocket. Successful solutions were deter-
mined based on the best ligand mode with RMSD ≤ 2.0 Å compared to the experi-
mental conformation. Then, four known molecules were taken from the CHEMBL
database and docked with NiV-G to assess their docking protocol accuracy. Then,
structure-optimized FDA-approved drugs were taken for screening against the NiV
attachment glycoprotein (NiV-G) utilizing the molecular docking method. Seven-
teen drugs were found to be potent inhibitors against Nipah virus and then narrowed
down to three novel inhibitors—nilotinib, acetyldigitoxin, and deslanoside following
topological analysis of chemical–protein interaction network, formed by integrating
drug–target network, human protein–protein interaction network, and NiV–human
interaction network. Both acetyldigitoxin and deslanoside were previously known
to be in the category of NiV inhibitors. In contrast, nilotinib is a part of benzanoids
class, which was previously not identified as NiV inhibitors [48].
5 Targeted Computational Approaches to Identify Potential Inhibitors … 153

5.8 Drug–Target–Drug Network-Based Approach

Few researchers also used drug–target–drug network-based approach like Rajput

et al., in 2020, who identified the repurposed drugs against 14 epidemic/pandemic
causing viruses, including NiV, through drug-target-drug network analysis. In this,
they manually extracted out drugs and their targets, which are already experimentally
validated either in vitro or in vivo. Then, these extracted drug targets were used for
fetching out new potent repurposed drugs. Then, prioritize the identified repurposed
drugs based on confidence score, i.e., the number of drug targets mapped in the
repurposed drugs divided by the total number of targets mapped to experimentally
validated drugs. Sixteen repurposed drugs were found shared between NiV and HeV.
Further, they performed the pathway analysis using the KEGGREST package in R/
Bioconductor, which showed that most drug targets participated in cancer signaling
pathways. Lastly, they performed molecular docking to validate and prioritize the
identified repurposed drugs [49].
In conclusion, we have recapitulated the experimentally tested antivirals studies
as well as the in silico approaches studies, which will be helpful for the researchers
in antiviral drug discovery against NiV.

References

1. Eaton BT, Broder CC, Middleton D, Wang LF (2006) Hendra and Nipah viruses: different and
dangerous. Nat Rev Microbiol 4(1):23–35. https://doi.org/10.1038/nrmicro1323
2. Pillai, V. S., Krishna, G., & Veettil, M. V. (2020). Nipah virus: past outbreaks and future
containment. In: Viruses, vol 12, Issue 4. MDPI AG. https://doi.org/10.3390/v12040465
3. Banerjee S, Gupta N, Kodan P, Mittal A, Ray Y, Nischal N, Soneja M, Biswas A, Wig N (2019)
Nipah virus disease: a rare and intractable disease. In: Intractable and rare diseases research,
vol 8, Issue 1, pp 1–8. International Advancement Center for Medicine and Health Research.
https://doi.org/10.5582/irdr.2018.01130
4. Aditi, Shariff M (2019) Nipah virus infection: a review. In: Epidemiology and infection, vol
147. Cambridge University Press. https://doi.org/10.1017/S0950268819000086
5. Skowron K, Bauza-Kaszewska J, Grudlewska-Buda K, Wiktorczyk-Kapischke N, Zacharski
M, Bernaciak Z, Gospodarek-Komkowska E (2022) Nipah virus–Another threat from the world
of zoonotic viruses. In: Frontiers in microbiology, vol 12. Frontiers Media S.A. https://doi.org/
10.3389/fmicb.2021.811157
6. Arunkumar G, Chandni R, Mourya DT, Singh SK, Sadanandan R, Sudan P, Bhargava B (2019)
Outbreak investigation of Nipah virus disease in Kerala, India, 2018. J Infect Dis 219(12):1867–
1878. https://doi.org/10.1093/infdis/jiy612
7. Singh RK, Dhama K, Chakraborty S, Tiwari R, Natesan S, Khandia R, Munjal A, Vora KS,
Latheef SK, Karthik K, Singh Malik Y, Singh R, Chaicumpa W, Mourya DT (2019) Nipah
virus: epidemiology, pathology, immunobiology and advances in diagnosis, vaccine designing
and control strategies—A comprehensive review. Veterinary Quarterly 39(1):26–55. https://
doi.org/10.1080/01652176.2019.1580827
8. Ksiazek TG, Rota PA, Rollin PE (2011) A review of Nipah and Hendra viruses with an historical
aside. In: Virus research, vol 162, issues 1–2, pp 173–183. https://doi.org/10.1016/j.virusres.
2011.09.026
154 S. Gautam and M. Kumar

9. Harcourt BH, Tamin A, Ksiazek TG, Rollin PE, Anderson LJ, Bellini WJ, Rota PA (2000)
Molecular characterization of Nipah virus, a newly emergent paramyxovirus. Virology
271(2):334–349. https://doi.org/10.1006/viro.2000.0340
10. Sun B, Jia L, Liang B, Chen Q, Liu D (2018) Phylogeography, transmission, and viral proteins
of Nipah virus. Virologica Sinica 33(5):385–393. https://doi.org/10.1007/s12250-018-0050-1
11. Ochani RK, Batra S, Shaikh A, Asad A (2019) Nipah virus the rising epidemic: a review.
Infezioni in Medicina 27(2):117–127
12. Lo MK, Rota PA (2008) The emergence of Nipah virus, a highly pathogenic paramyxovirus.
J Clin Virol 43(4):396–400. https://doi.org/10.1016/j.jcv.2008.08.007
13. Ashburn TT, Thor KB (2004) Drug repositioning: identifying and developing new uses for
existing drugs. Nat Rev Drug Discovery 3(8):673–683. https://doi.org/10.1038/nrd1468
14. Pushpakom S, Iorio F, Eyers PA, Escott KJ, Hopper S, Wells A, Doig A, Guilliams T, Latimer
J, McNamee C, Norris A, Sanseau P, Cavalla D, Pirmohamed M (2018) Drug repurposing:
progress, challenges and recommendations. In: Nature reviews drug discovery, Vol. 18, issue
1. Nature Publishing Group, pp 41–58. https://doi.org/10.1038/nrd.2018.168
15. Chong HT, Kamarulzaman A, Tan CT, Goh KJ, Thayaparan T, Kunjapan SR, Chew NK,
Chua KB, Lam SK (2001) Treatment of acute Nipah encephalitis with ribavirin. Ann Neurol
49(6):810–813. https://doi.org/10.1002/ana.1062
16. Georges-Courbot MC, Contamin H, Faure C, Loth P, Baize S, Leyssen P, Neyts J, Deubel V
(2006) Poly(I)-poly(C12U) but not ribavirin prevents death in a hamster model of Nipah virus
infection. Antimicrob Agents Chemother 50(5):1768–1772. https://doi.org/10.1128/AAC.50.
5.1768-1772.2006
17. Aljofan M, Sganga ML, Lo MK, Rootes CL, Porotto M, Meyer AG, Saubern S, Moscona A,
Mungall BA (2009) Antiviral activity of gliotoxin, gentian violet and brilliant green against
Nipah and Hendra virus in vitro. Virol J 6:1–13. https://doi.org/10.1186/1743-422X-6-187
18. Pallister J, Middleton D, Crameri G, Yamada M, Klein R, Hancock TJ, Foord A, Shiell B,
Michalski W, Broder CC, Wang L-F (2009) Chloroquine administration does not prevent Nipah
virus infection and disease in ferrets. J Virol 83(22):11979–11982. https://doi.org/10.1128/jvi.
01847-09
19. Freiberg AN, Worthy MN, Lee B, Holbrook MR (2010) Combined chloroquine and ribavirin
treatment does not prevent death in a hamster model of Nipah and Hendra virus infection. J
Gen Virol 91(3):765–772. https://doi.org/10.1099/vir.0.017269-0
20. Mohr EL, McMullan LK, Lo MK, Spengler JR, Bergeron É, Albariño CG, Shrivastava-Ranjan
P, Chiang CF, Nichol ST, Spiropoulou CF, Flint M (2015) Inhibitors of cellular kinases with
broad-spectrum antiviral activity for hemorrhagic fever viruses. Antiviral Res 120:40–47.
https://doi.org/10.1016/j.antiviral.2015.05.003
21. Hotard AL, He B, Nichol ST, Spiropoulou CF, Lo MK (2017) 4' -Azidocytidine (R1479) inhibits
henipaviruses and other paramyxoviruses with high potency. Antiviral Res 144:147–152. https:/
/doi.org/10.1016/j.antiviral.2017.06.011
22. Dawes BE, Kalveram B, Ikegami T, Juelich T, Smith JK, Zhang L, Park A, Lee B, Komeno
T, Furuta Y, Freiberg AN (2018). Favipiravir (T-705) protects against Nipah virus infection in
the hamster model /631/326/22/1295 /631/326/596/1296 /13/106 /14/35 /38/77 /82/51 /96/63
article. Sci Rep 8(1). https://doi.org/10.1038/s41598-018-25780-3
23. Lo MK, Feldmann F, Gary JM, Jordan R, Bannister R, Cronin J, Patel NR, Klena JD, Nichol
ST, Cihlar T, Zaki SR, Feldmann H, Spiropoulou CF, De Wit E (2019) Remdesivir (GS-5734)
protects African green monkeys from Nipah virus challenge. Sci Transl Med 11(494). https://
doi.org/10.1126/scitranslmed.aau9242
24. Lo MK, Amblard F, Flint M, Chatterjee P, Kasthuri M, Li C, Russell O, Verma K, Bassit L,
Schinazi RF, Nichol ST, Spiropoulou CF (2020) Potent in vitro activity of β-D-4' -chloromethyl-
2' -deoxy-2' -fluorocytidine against Nipah virus. Antiviral Res 175:104712. https://doi.org/10.
1016/j.antiviral.2020.104712
25. Zhu Z, Dimitrov AS, Bossart KN, Crameri G, Bishop KA, Choudhry V, Mungall BA, Feng
Y-R, Choudhary A, Zhang M-Y, Feng Y, Wang L-F, Xiao X, Eaton BT, Broder CC, Dimitrov
DS (2006) Potent neutralization of Hendra and Nipah viruses by human monoclonal antibodies.
J Virol 80(2):891–899. https://doi.org/10.1128/jvi.80.2.891-899.2006
5 Targeted Computational Approaches to Identify Potential Inhibitors … 155

26. Dang HV, Chan YP, Park YJ, Snijder J, Da Silva SC, Vu B, Yan L, Feng YR, Rockx B, Geisbert
TW, Mire CE, Broder CC, Veesler D (2019) An antibody against the F glycoprotein inhibits
Nipah and Hendra virus infections. Nat Struct Mol Biol 26(10):980–987. https://doi.org/10.
1038/s41594-019-0308-9
27. Dang HV, Cross RW, Borisevich V, Bornholdt ZA, West BR, Chan YP, Mire CE, Da Silva
SC, Dimitrov AS, Yan L, Amaya M, Navaratnarajah CK, Zeitlin L, Geisbert TW, Broder CC,
Veesler D (2021) Broadly neutralizing antibody cocktails targeting Nipah virus and Hendra
virus fusion glycoproteins. Nat Struct Mol Biol 28(5):426–434. https://doi.org/10.1038/s41
594-021-00584-8
28. Mathieu C, Porotto M, Figueira TN, Horvat B, Moscona A (2018) Fusion inhibitory lipopep-
tides engineered for prophylaxis of Nipah virus in primates. J Infect Dis 218(2):218–227. https:/
/doi.org/10.1093/infdis/jiy152
29. Mungall BA, Schopman NCT, Lambeth LS, Doran TJ (2008) Inhibition of Henipavirus infec-
tion by RNA interference. Antiviral Res 80(3):324–331. https://doi.org/10.1016/j.antiviral.
2008.07.004
30. Rajput A, Kumar M (2022) Anti-Ebola: an initiative to predict Ebola virus inhibitors through
machine learning. Mol Diversity 26(3):1635–1644. https://doi.org/10.1007/s11030-021-102
91-7
31. Rajput A, Thakur A, Mukhopadhyay A, Kamboj S, Rastogi A, Gautam S, Jassal H, Kumar
M (2021) Prediction of repurposed drugs for Coronaviruses using artificial intelligence and
machine learning. Comput Struct Biotechnol J 19:3133–3148. https://doi.org/10.1016/j.csbj.
2021.05.037
32. Qureshi A, Rajput A, Kaur G, Kumar M (2018) HIVprotI: an integrated web based platform
for prediction and design of HIV proteins inhibitors. J Cheminf 10(1). https://doi.org/10.1186/
s13321-018-0266-y
33. Rajput A, Kumar M (2018) Anti-flavi: a web platform to predict inhibitors of flaviviruses using
QSAR and peptidomimetic approaches. Front Microbiol 9:3121. https://doi.org/10.3389/fmicb.
2018.03121
34. Qureshi A, Kaur G, Kumar M (2017) AVCpred: an integrated web server for prediction and
design of antiviral compounds. Chem Biol Drug Des 89(1):74–83. https://doi.org/10.1111/
cbdd.12834
35. Qureshi A, Tandon H, Kumar M (2015) AVP-IC50Pred: multiple machine learning techniques-
based prediction of peptide antiviral activity in terms of half maximal inhibitory concentration
(IC50). Biopolymers 104(6):753–763. https://doi.org/10.1002/bip.22703
36. Thakur N, Qureshi A, Kumar M (2012) AVPpred: collection and prediction of highly effective
antiviral peptides. Nucl Acids Res 40(W1). https://doi.org/10.1093/nar/gks450
37. Rajput A, Kumar A, Megha K, Thakur A, Kumar M (2021) DrugRepV: a compendium of
repurposed drugs and chemicals targeting epidemic and pandemic viruses. Brief Bioinform
22(2):1076–1084. https://doi.org/10.1093/bib/bbaa421
38. Qureshi A, Thakur N, Tandon H, Kumar M (2014) AVPdb: a database of experimentally
validated antiviral peptides targeting medically important viruses. Nucl Acids Res 42(D1).
https://doi.org/10.1093/nar/gkt1191
39. Rajput A, Kumar A, Kumar M (2019) Computational identification of inhibitors using QSAR
approach against Nipah virus. Front Pharmacol 10(FEB). https://doi.org/10.3389/fphar.2019.
00071
40. Dar AM, Mir S (2017) Molecular docking: approaches, types, applications and basic challenges.
J Anal Bioanal Tech 08(02):8–10. https://doi.org/10.4172/2155-9872.1000356
41. Lipin R, Dhanabalan AK, Gunasekaran K, Solomon RV (2021) Piperazine-substituted deriva-
tives of favipiravir for Nipah virus inhibition: What do in silico studies unravel? SN Appl Sci
3(1). https://doi.org/10.1007/s42452-020-04051-9
42. James JP, Apoorva, Monteiro SR., Sukesh KB, Varun A (2021) Design and identification of
lead compounds targeting Nipah G attachment glycoprotein by in silico approaches. J Pharm
Res Int 156–169. https://doi.org/10.9734/jpri/2021/v33i40a32232
156 S. Gautam and M. Kumar

43. Sen N, Kanitkar TR, Roy AA, Soni N, Amritkar K, Supekar S, Nair S, Singh G, Madhusudhan
MS (2019) Predicting and designing therapeutics against the Nipah virus. PLoS Negl Trop Dis
13(12):e0007419. https://doi.org/10.1371/journal.pntd.0007419
44. Ropón-Palacios G, Chenet-Zuta ME, Olivos-Ramirez GE, Otazu K, Acurio-Saavedra J, Camps
I (2020) Potential novel inhibitors against emerging zoonotic pathogen Nipah virus: a virtual
screening and molecular dynamics approach. J Biomol Struct Dyn 38(11):3225–3234. https://
doi.org/10.1080/07391102.2019.1655480
45. Kalbhor MS, Bhowmick S, Alanazi AM, Patil PC, Islam MA (2021) Multi-step molecular
docking and dynamics simulation-based screening of large antiviral specific chemical libraries
for identification of Nipah virus glycoprotein inhibitors. Biophys Chem 270:106537. https://
doi.org/10.1016/j.bpc.2020.106537
46. Ahmed Bhuiyan M, Atia Keya N, Susan Mou F, Rahman Imon R, Alam R, Ahammad F (2020)
Discovery of potential compounds against nipah virus: a molecular docking and dynamics
simulation approaches. March. https://doi.org/10.21203/rs.3.rs-1398424/v1
47. Randhawa V, Pathania S, Kumar M (2022) Computational identification of potential multitarget
inhibitors of Nipah virus by molecular docking and molecular dynamics. Microorganisms
10(6):1181. https://doi.org/10.3390/microorganisms10061181
48. Pathania S, Randhawa V, Kumar M (2020) Identifying potential entry inhibitors for emerging
Nipah virus by molecular docking and chemical-protein interaction network. J Biomol Struct
Dyn 38(17):5108–5125. https://doi.org/10.1080/07391102.2019.1696705
49. Rajput A, Thakur A, Rastogi A, Choudhury S, Kumar M (2021) Computational identification
of repurposed drugs against viruses causing epidemics and pandemics via drug-target network
analysis. Comput Biol Med 136:104677. https://doi.org/10.1016/j.compbiomed.2021.104677
Chapter 6
Role of Computational Modelling
in Drug Discovery for HIV

Anish Gomatam, Afreen Khan, Kavita Raikuvar, Merwyn D’costa,

and Evans Coutinho

Abstract With over 36 million people currently living with HIV, HIV/AIDS
continues to have devastating effects on human health worldwide. Viral resistance
to anti-HIV drugs remains a major cause of concern, necessitating a regimen of
highly active antiretroviral therapy (HAART), which consists of a combination of
multiple drugs for long-term clinical benefit. Clearly, the rapid development of novel
molecules that can help change the present regimen to new drug combinations is crit-
ical for tackling the resistance problem. In this regard, computational methods have
emerged as a valuable tool in HIV research, contributing greatly to our understanding
of HIV biology and aiding in the design of potent anti-HIV compounds. This chapter
gives an overview of the various computational strategies reported in the discovery of
drugs for the treatment of HIV. A comprehensive overview of several structure-based
and ligand-based computational methods is presented first; this is followed by some
notable applications of these methods in the discovery of novel anti-HIV compounds.
Finally, we discuss the emergence of powerful machine learning algorithms which
have proven useful both in the design of new compounds and in the development of
theoretical models that can predict resistance to antiretroviral therapy.

Keywords AIDS · Computational · HAART · HIV · Modelling

6.1 Background

Despite significant endeavours and treatment advancements since the Pasteur Insti-
tute in France isolated and identified the human immunodeficiency virus-1 (HIV-1),
HIV has been a serious worldwide health threat [1]. Globally in 2020, around 37.7
million people were living with HIV, this number comprises 36 million adults and
1.7 million children under the age of 15. There were also 1.5 million new infec-
tions in 2020 with over 680,000 fatalities [2]. HIV belongs to the genus lentivirus

A. Gomatam · A. Khan · K. Raikuvar · M. D’costa · E. Coutinho (B)

Department of Pharmaceutical Chemistry, Bombay College of Pharmacy, Mumbai, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 157
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_6
158 A. Gomatam et al.

within the retrovirus family [3]. Unlike some retroviruses, HIV is not capable of self-
replication. HIV targets the body’s immune system by attacking the CD4 T-cells over
time in order to multiply and proliferate throughout the body; this directly impacts the
immunity of the host, rendering the host more prone to opportunistic infections. In
many cases, HIV progresses to its most advanced form, acquired immune deficiency
syndrome (AIDS) if left untreated within 10 years, during which deadly infections
and malignancies are common [4]. HIV usually spreads through sexual contact or by
the transmission of blood, pre-ejaculate, sperm, and vaginal secretions. A pregnant
HIV-positive woman may pass her infection to her unborn child through breast milk,
blood, or vaginal discharge. An infected person may not exhibit any symptoms or
might go through a brief period of influenza-like sickness. As the infection affects
the immune system, physical weakness and enlargement of the neck lymph nodes are
commonly observed, and in cases where adequate treatment is not given, life threat-
ening conditions such as tuberculosis, severe bacterial infections, and even cancers
such as lymphomas may occur [2, 5].

6.2 HIV Replication Cycle

The HIV-1 life cycle is complex and may be divided into two phases: early replication
and late replication. The virion’s adhesion to the T-cell surface and the proviral DNA’s
incorporation into the host genome are the first two stages of the early phase [6].
Proviral transcription begins at the late phase of replication and continues until fully
infectious offspring virions are produced [7]. The various stages in the lifecycle of
HIV and drug targets are depicted in Fig. 6.1 and are briefly discussed below [7–9].

1. Binding: The first step of viral attack is characterized by attachment of the virus
to the T-cell surface through CD4 or CXCR4 or CCR5 receptors. This causes the
binding of gp120, a monomeric virion associated protein, to form a gp120-CD4
complex.
2. Fusion: The viral envelope undergoes structural changes after adhering to the
CD4 cell, causing the virus to fuse with the cell membrane resulting in bursting
of the viral envelope. After entering the T-cell, the virus releases its RNA along
with the enzymes—reverse transcriptase and integrase.
3. Reverse transcription: The enzyme reverse transcriptase converts HIV single
strand RNA (ssRNA) to HIV double strand DNA (dsDNA), which allows it
to enter the nucleus and combine with the cell’s genetic material, resulting in
virulent activity.
4. Integration: At this stage of the HIV life cycle, the enzyme integrase inserts
newly transcribed viral DNA into the host DNA, causing the cell to become
virulent.
6 Role of Computational Modelling in Drug Discovery for HIV 159

Fig. 6.1 Illustration of the HIV life cycle

160 A. Gomatam et al.

5. Replication: After the viral DNA is integrated with the host DNA by integrase,
the reproduction process of the virus begins. This process occurs when the virus
begins to replicate or produce long chain HIV proteins using the machinery of
the host cell.
6. Assembly: The sixth stage of the HIV lifecycle is the most crucial stage since it
is here that the virus begins assembly of the components after manufacturing the
essential components in the fifth stage. The new HIV RNA along with essential
HIV precursor proteins generated by the host CD4 cells are ferried to the cell
surface during this step, where the components are combined into a structurally
complete but immature non-infectious virus.
7. Budding: As HIV pushes itself out of the host cell, it remains non-infectious.
The HIV lifecycle ends with the production of mature infectious virions resulting
from the action of protease which breaks down the immature virus’s long protein
chains, transforming it into a mature virus capable of infecting other healthy
cells.

Since the introduction of the first antiretroviral therapy (ART), zidovudine

(AZT)—a nucleoside reverse transcriptase inhibitor (NRTI), in 1987 [10], HIV treat-
ments have advanced significantly. Highly active antiretroviral therapy (HAART)
regimens, also known as combinatorial ART (cART), is now the primary treatment
for HIV [11]. Antiretroviral drugs can assist in lowering viral load, fight infections,
and enhance the overall quality of life. Anti-HIV medications are classified into
six categories: nucleoside or nucleotide reverse transcriptase inhibitors (NRTIs),
non-nucleoside reverse transcriptase inhibitors (NNRTIs), protease inhibitors (PIs),
integrase inhibitors, fusion inhibitors, and co-receptor inhibitors. We provide a clas-
sification of the anti-HIV medications in Table 6.1 [10, 12]. Recent efforts towards
development of in HIV-1 reverse transcriptase inhibitors, which are in the clinical
phase are summarized in the paper by Shaung et al. [13].

6.3 The Resistance Problem

Despite the progress made in antiretroviral therapy, the management of HIV has
been hindered greatly by the emergence of resistance. When the virus is targeted by
antiretroviral agents, it may undergo a change in its genetic material, i.e. the naturally
occurring or ‘wild type’ version of the genome. This is referred to as a ‘mutation’
and may lead to an inability of the drug to block the replication of the virus, thus
causing the virus to become ‘resistant’ to the drug. Due to the advent of drug-
resistant viruses, all antiretroviral medications, including those from more recent
pharmacological classes, are at danger of becoming partly or completely ineffective
[14].
Several aspects associated with the HIV life cycle and replication are important
contributors to the organism’s fast and widespread establishment of resistance. The
HIV-reverse transcriptase (RT) enzyme is known for its ‘poor fidelity’ (i.e. the enzyme
Table 6.1 Summary of approved drugs for treatment of HIV encompassing the five major classes
Nucleoside or -tide reverse Non-nucleoside Protease inhibitors (PIs) Integrase strand Entry inhibitors
transcriptase inhibitors reverse transfer inhibitors Fusion inhibitors Co-receptor
(NRTIs) transcriptase (INSTIs) inhibitors
inhibitors
(NNRTIs)
Mechanism Competitive inhibition of Alter HIV-RT Competitively block the Prevent integration Prevent viral entry into host cell. Fusion
of action nucleic acid from viruses by structure by protease enzyme by of the viral DNA inhibitors prevent fusion of the virus
causing chain termination binding to an binding to its catalytic into the host with the host cell, whereas CCR5
during reverse transcription allosteric region site with a high affinity genome antagonists prevent infection of the CD4
about 10 Å away and inhibiting the T-cells by blocking the CCR5 receptor
from the enzyme’s enzyme’s ability to
active site, thus function, resulting in the
preventing reverse production of immature
transcription and non-contagious viral
particles
FDA Nucleoside Nucleotide Nevirapine Saquinavir-1995 Raltegravir-2007 E-Fostemsavir-2020 Maraviroc-2007
approved Zidovudine Tenofovir (NVP)-1996 Ritonavir-1996 Dolutegravir-2013
drugs (AZT)-1987 disoproxil Delavirdine Indinavir-1996 Elvitegravir-2014
Didanosine fumarate (DLV)-1997 Nelfinavir-1997 Cabotegravir-2021
6 Role of Computational Modelling in Drug Discovery for HIV

(ddI)-1991 (TDF)-2001 Efavirenz Amprenavir-1999

Zalcitabine (EFV)-1998 Atazanavir
(ddC)-1992 Etravirine Fosamprenavir-2003
Stavudine (ETV)-2008 Tipranavir-2005
(d4T)-1994 Rilpivirine Darunavir-2006
Lamivudine (RPV)-2011
(3TC)-1995 Elsulfavirine
Abacavir (ESV)–2017
(ABC)-1998 Doravirine
Emtricitabine (DOR)-2018
(FTC)-2003
161
162 A. Gomatam et al.

is rather nonselective throughout the copying process) and is prone to introducing

errors while transcribing viral RNA into DNA [11, 15]. According to some estimates,
HIV-RT introduces one mutation for every viral genome that is transcribed. The high
mutation rate of HIV-RT, when combined with the high levels of viral production
and turnover, means that the patient may oftentimes have a diverse mixture of viral
quasi species within a few weeks of infection. One or more of these viral quasi
species may be resistant to medication, and the quasi species which confer the most
advantage to the virus (i.e. reducing susceptibility to an antiviral agent) are retained
through a process of Darwinian selection. Resistance may also develop independently
of ART therapy, this happens when an individual contracts HIV for the first time
from a resistant strain, usually transmitted from a HIV-positive person undergoing
antiretroviral therapy. This is referred to as primary or acquired resistance. Significant
advancements have made been in successfully identifying mutations linked to drug
resistance and comprehending the processes through which they confer resistance.
Several mechanisms have been discovered, and they vary for drugs both from the
same class and from other classes [14, 16]. These have been covered comprehensively
in reviews by Cilento et al. [11] and Collier et al. [17]
Clearly, we now face an urgent problem, as the number of persons with HIV
resistant strains is rising. To combat medication resistance and reduce the high costs
associated with treatment, new drug classes must continuously be investigated and
developed.

6.4 Structure-Based Methods

In silico drug design or computer-aided drug design (CADD) has the potential
to accelerate the tedious process of designing and developing a drug candidate.
With the recent developments in the architecture and algorithms of structure-based
drug design, intensive computations can be performed in a time-affordable manner
[18]. The structure-based drug design (SBDD) techniques require an experimen-
tally solved structure of the protein or better still the protein–ligand complex,
which is obtained using X-ray crystallography, nuclear magnetic resonance (NMR)
spectroscopy, and comparative protein modelling or homology modelling.
The SBDD-based methods help not only in identification of hits but also in the
ability to optimize them. The docking method, which is central to SBDD, is based
on the theory of the lock and key mechanism, where both the protein and ligand are
considered as rigid bodies. The introduction of the ‘induced-fit’ theory, proposed by
Koshland gave us an understanding that the active site of the protein continuously
undergoes changes as the ligand interacts with the protein [19]. Ideally, both the
ligand and protein should be considered flexible bodies. However, keeping in mind
the limitations of computational resources, the protein may often be treated as a rigid
body. There are multiple algorithms [20] that have been devised to treat the docking
problem. To summarize, molecular docking methods are useful in predicting the
binding pose of the ligand (drug) to the protein and its relative affinity [21]. Binding
6 Role of Computational Modelling in Drug Discovery for HIV 163

Fig. 6.2 Examples of HIV drugs currently on the market discovered using SBDD methods
[saquinavir-SQV, ritonavir-RTV, and indinavir-INV]

studies using molecular dynamics (MD) simulations [22] on the ‘best’ binding pose
can be performed to understand the affinity. MD can help in ligand optimization and
provide an insight on the pathways and kinetics of interactions. Virtual screening
(VS) is a technique where docking is carried out on a large dataset of molecules to
identify potential leads among them. The HIV drugs that have been discovered using
the SBDD techniques [23] are given in Fig. 6.2.
The structure-based drug design methodology has been implemented and proved
to hasten the drug discovery process in several instances. Recent developments and
directions in molecular docking and MD simulations are discussed below.

6.4.1 Molecular Docking

Molecular docking is one of the most widely used in silico techniques for drug design.
A set of ligands with optimized geometry is set to position in the binding site of the
protein, i.e. to generate a pose of the ligand with a corresponding binding energy
[24]. The method is often coupled to a scoring function to calculate a score, which
reflects the ‘goodness of fit’ of the ligand pose in the protein pocket. The best pose is
identified as the one making optimal contacts with the amino acids in the active site,
as reported experimentally. To predict the binding of ligands with a mutant protein, a
molecular modelling software [25] can be used to mutate specific residues (most of
the times the ones that are reported experimentally). The major source for 3D protein
and protein–ligand structures is the Protein Data Bank (PDB, https://www.rcsb.org).
The outline of the molecular docking process is shown in Fig. 6.3.
164 A. Gomatam et al.

Fig. 6.3 Schematic representation of the molecular docking technique

The molecular docking technique is initially used to understand the interactions

between the protein and experimentally confirmed inhibitor. This step acts as a vali-
dation of the methodology adopted and is followed by virtual screening, so as to find
new inhibitors specific to the target. The technique has also been used to explain the
resistance problem [26]. The interactions that are absent in the mutated protein can
be identified; this can be exploited to design drugs acting on the resistant strains. The
docking programmes most commonly used are AutoDock 4.0 [27], Sybyl-X [28],
AutoDock Vina [29], Molegro Virtual Docking [30], AutoDock Racoon [31], and
Glide (Schrodinger Inc.) [32]. We now focus on the application of molecular docking
as applied to the discovery of HIV-1 inhibitors.
Herbal compounds have paved a way for identification of many drugs, and many
researchers have made efforts to identify anti-HIV agents from natural sources. Vora
et al. [33] have identified five natural compounds against HIV. These are anolignan B
(active against reverse transcriptase), curcumin (an inhibitor of the integrase enzyme),
mulberroside C (protease inhibitor), chebulic acid (ribonuclease inhibitor), and neo-
andrographolide (entry inhibitor). Very few of the natural compounds were found
to have an IC50 value that is close to the synthetic drugs, so chemical modifications
are needed to improve their activity, nevertheless, these could be considered in near
future as leads.
A review by Tarasova et al. [34] provides an outline of the molecular docking
technique applied to understand structural changes in reverse transcriptase (RT)
6 Role of Computational Modelling in Drug Discovery for HIV 165

associated with HIV-1 resistance. Most of the docking studies reported in litera-
ture are performed on the NNRTIs, rather than on the NRTIs. This is attributed to
two reasons—first, the NRTIs act via a competitive mechanism and second, the esti-
mation of the binding energy between the NRTIs and the protein is difficult. Also,
that there are several possible mechanisms that exist for the resistance, and the role
of each mutation in the level of resistance is not known. Perhaps, the discovery of
the mechanisms of HIV-1 resistance as revealed by Tarasova et. al. can pave ways
for molecular docking applications to develop highly active NRTIs. A compilation
of the studies discussed here is presented in Table 6.2.
Singh et al. [35] have explored diarylpyrimidine derivatives as NNRTIs. They
docked the compounds at the allosteric site of HIV-RT and identified eight potential
ligands having a profile better than the known inhibitor etravirine. Further, molecular
dynamics and free energy-based calculations were done to understand the binding
affinity and stability of the protein–ligand complexes. Compound 6 in the study
showed better stability and inhibition than the reference drug, paving the way for the
development of newer second generation NNRTIs. In a study reported by Fraczek
et al. [36], a comparison of different molecular docking techniques on a set of poten-
tial NNRTIs was carried out. The paper describes a comparison of FlexX, Hyde,
Molegro Virtual Docker, Glide, and AutoDock Vina on their ability to predict RT
inhibitory activity of 1,2,4-triazoles (n = 111) and azoles (n = 76) as NNRTIs. They
showed that the correlation between the experimentally determined half maximal

Table 6.2 Summary of the docking studies discussed

Target PDB ID Ligand Software References
Protease, 5KAO, Natural products Discovery [33]
Ribonuclease, 3QIN, Studio,
IN, RT, gp-120 5EU7, Schrodinger,
4G1Q, Molegro Virtual
1G9M Docking
HIV RT 3MEC Diarylpyrimidine derivatives Discovery [35]
Studio 3.0
HIV RT 3DLG 1,2,4-triazole and azole Glide, FlexX, [36]
derivatives Molegro Virtual
Docker,
AutoDock Vina,
Hyde, Sybyl-X
HIV RT 3MEC Diarylpyrimidine derivatives Sybyl-X [37]
HIV RT 4G1Q Amino-oxy-diarylquinoline AutoDock 4.2 [38]
HIV RT 1S1U Cu (II) ion Schiff base AutoDock Vina [39]
complexes
HIV RNase 5J1E Hydroxypyrimidine-2,4-diones Sybyl-X 2.1 [40]
HIV IN 3OYA Benzoxazoline, quinazoline, AutoDock Vina [41]
diazocoumarin
[HIV-RT: HIV-reverse transcriptase, IN: integrase]
166 A. Gomatam et al.

effective concentration (EC50 ) and predicted binding energies were highly dependent
on the ligand set. The performance of all the docking programmes was comparable;
however, AutoDock Vina, Molegro Virtual Docker, and Hyde indicate that shape
matching of ligand and binding sites is the preferred method for identifying inhibitors.
However, activity prediction was restricted to only those substances closely akin to
a natural ligand. A study on compounds containing the diarylpyrimidine core was
reported by Liu et al. [37] as NNRTIs. They performed molecular docking using the
Sybyl-X software to generate the 3D binding pose; these structures were then used to
calculate various 3D descriptors from which a 3D-QSAR model was built using the
comparative molecular field analysis (CoMFA) and comparative molecular similarity
indices analysis (CoMSIA) approaches. The phenyl group present in the diarylpyrim-
idines (Fig. 6.4) was able to engage in π-π stacking interaction with the aromatic
residues of the binding site; in contrast, the cycloalkanes are unable to do so. This
shows that the phenyl ring at the C4-position of the pyrimidine ring is preferred over
cycloalkane motifs for good activity. The best ligands for the 3D-QSAR models were
found to be substituents having a 4-isopropyl, 3-hydroxy, 2-fluoro-4-methyl groups.
Makarasen et al. [38] designed derivatives containing the amino-oxy-diarylquinoline
core as NNRTIs from a pharmacophore model that was constructed on the interaction
templates of nevirapine, efavirenz, etravirine, and rilpivirine. Also using molecular
docking, they were able to identify important interactions of the ligands with Lys101
and His235 residues via a hydrogen bond and with Tyr318 via π-π stacking. These
compounds were synthesized and tested and found to have an inhibition rate of about
39.7% at 1 μM concentration. Shanty et al. [39] identified Cu(II) ion complexes with
heterocyclic Schiff bases as inhibitors of HIV-1 RT. These molecules were docked
against the protein. The paper summarizes the different types of complexes and their
stability towards binding and points out that hydrogen bonding, hydrophobic inter-
actions, and the π-sulphur contacts are crucial interactions. The compounds were
synthesized, tested with nevirapine as the reference drug, and were found to be active
with an inhibition rate of 86 versus 100% for nevirapine against the target enzyme.
For HIV multiplication, the HIV-1 RT-ribonuclease H (RNase-H) association
plays an important role as reported by Gao et al. [40]. In this study, a series of
hydroxypyrimidine-2,4-diones (n = 93) was curated, and in silico methods like
docking followed by MD simulations were performed. The final poses were used to
calculate the various physicochemical descriptors, and a 3D-QSAR model was built.
For the CoMFA model, the validation metrics are r 2 0.949, q2 0.908, and F value
of 492.826. The steric and electrostatic field contributions are 72.0% and 28.0%,
respectively, showing that the steric field contributes more to activity according to
the CoMFA model. To sum up, the following substituents may be introduced into
appropriate areas to enhance the inhibitory activity of hydroxypyrimidine-2,4-diones:
the pyrimidine ring’s N1 position is positively charged and can be substituted by small
groups; the N3 position is negatively charged and can be substituted by hydrophilic
substituents; the linker moiety can be attached with hydrophobic groups; the 2nd or
3rd position of the aromatic moiety can be substituted by bulky, negatively charged,
and/or hydrophobic groups; the 2' or 4' position of the aromatic moiety can carry
bulky, negatively charged, and/or hydrophobic groups; and the 3' position of the
6 Role of Computational Modelling in Drug Discovery for HIV 167

Fig. 6.4 Structure of the ligands—compound 6 [28], 35 [30], 19 [34], and the diarylpyrimidine
core [33]

aromatic moiety can accommodate negatively charged groups. The newly designed
molecules were proposed as leads for HIV RNase-H inhibitors. An investigation on
chalcone derivatives as HIV-1 protease inhibitors was reported by Turkovic et al.
[42]. They curated a set of 20 structurally similar chalcones and docked them in the
protease enzyme, to decipher the interactions. These molecules were synthesized and
tested for anti-HIV-1 activity via a fluorimetric assay. The best molecule exhibited an
IC50 of 0.001 μM, which is comparable to the commercially available drug Darunavir.
Novel 2,3-diaryl-4-quinazolinone derivatives were designed by Hajimahdi et al. [43];
they were docked in the HIV-1 integrase enzyme, and the top ranking molecules
were synthesized and assayed for their anti-HIV activity. The study provided novel
leads, with the best molecule showing an EC50 of 37 μM. Kamyar et al. [41] have
explored quinazoline, benzoxazolinone, and diazocoumarin derivatives as anti-HIV
agents. The study describes a set of 29 compounds which were docked against the
HIV integrase protein using the AutoDock Vina package. Compound 19 in the set
binds to the active site of integrase by two major moieties—first is the carbonyl
group of the compound which binds to the Mg2+ ions and the second is the aryl
side chain which fits into the hydrophobic pocket at the protein-DNA interface via
a π-stacking interaction. This docking data could provide useful insights for design
of new anti-HIV agents.
168 A. Gomatam et al.

6.4.2 Molecular Dynamics and Free Energy Calculations

Classical MD by its very nature is able to account for the structural flexibility of
the drug-protein system, which is well supported by the induced-fit and the confor-
mational selection theories [44]. This method is a physical model to understand the
interactions and motion of the atoms in a molecule as governed by Newton’s laws of
motion. Generally, a force field is applied to all the atoms present in the system and
this is used to estimate the overall energy of the system. When performing an MD
simulation, the integration of the laws of motion generates a series of configurations,
showcasing a trajectory that provides two crucial pieces of information—positions
and velocities of the atoms over time. This is used to calculate the free energies which
are correlated with the experimental observations to draw out conclusions about the
drug binding process with the target protein (Fig. 6.5).
The core idea of molecular dynamics is the study of the time-dependent behaviour
of the system. This is explained by Newton’s second law of motion:

∂ V [ri (t)]
f i (t) = m i ai (t) = − , (6.1)
∂ri (t)

where m i is the mass, ai (t) the acceleration, f i (t) the total force operating at a certain
moment in time t on the ith atom of the system. The vector r i (t) which depicts the
positions of the N interacting atoms in Cartesian space (r = {x 1 , y1 , z1; x 2 , y2 , z2; ….
x N , yN , zN }) represents the configuration of the system at the given instant.
The empirical potential energy equation is given by

Fig. 6.5 Schematic representation of the MD simulation workflow

6 Role of Computational Modelling in Drug Discovery for HIV 169
∑ ( )2 ∑ ( )2
E total = K r r − req + K ϑ ϑ − ϑeq
bonds angles
[ ]
∑ Vo[ ] ∑ Ai j Bi j qi q j
+ 1 + cos(nφ − γ ) + − 2 + . (6.2)
dihedrals
2 i< j
Ri2j Ri j ε Ri j

Equation 6.2 comprises of all forces arising from interactions between the bonded
and the non-bonded atoms. The bonded interactions include bonds, angles and dihe-
drals, while the non-bonded forces are those that arise due to the van der Waals
interactions depicted through the Lennard–Jones 6–12 potential and the Coulombic
electrostatic forces. These energy terms are parameterized to reproduce the real
behaviour of the molecules and are collectively called as the ‘force field’. The force
fields commonly used in MD simulations are the General Amber Force Field (GAFF)
[45], CHARMM [46], and GROMOS [47]. The positions of these atoms are moved
according to Newton’s laws of motion, using the calculated forces. The time step
in the MD simulation is a few (1 or 2) femtoseconds, and the process is repeated
several million times which gives the length of the simulation. The most popular
software packages for MD simulations are AMBER [48], NAMD [49], CHARMM
[50], and GROMACS [51]. Table 6.3 summarizes the different MD methods used
on the different targets of HIV discussed here.
The free energy calculations methods can be classified as the end-state free energy
methods, also called partitioning-based methods, and the non-partitioning-based
methods. The latter group of methods are generally more accurate and computa-
tionally exhaustive than the end-state free energy methods. Moreover, end-state free
energy methods allow the energy components to be decomposed into electrostatics,
van der Waals and bonded energy terms. The philosophy of the non-partitioning-
based methods denounces the idea that the free energy can be decomposed into
components. The molecular mechanics Poisson-Boltzmann surface area (MM-PB/
SA) and molecular mechanics generalized Born surface area (MM-GB/SA) methods
are mostly used and belong to the class of end-state free energy methods. Free energy
perturbation (FEP) and thermodynamic integration (TI) belong to the class of non-
partitioning-based methods. Furthermore, FEP and TI can be used to calculate abso-
lute as well as relative binding free energy, whereas MM-PB/SA and MM-GB/SA
methods yield only relative binding free energy [56].

Table 6.3 Summary of the MD simulation-based studies discussed in this section

Target PDB Method Software References
RNase-H 5J1E Unbiased all atom MD simulation GROMACS [40]
NNRTI 3DLG Unbiased all atom MD simulation GROMACS [52]
HIV protease 1HHP Unbiased all atom MD simulation AMBER12 [53]
HIV protease 1HHP Gaussian accelerated molecular dynamics AMBER14 [54]
(GaMD)
HIV integrase 6C0J Unbiased all atom MD simulation GROMACS [55]
170 A. Gomatam et al.

MM-PB/SA and MM-GB/SA methods calculate the free energy of binding by

adding a correction for solvation electrostatics to the molecular mechanics gas phase
energies. These solvation electrostatics are computed either by Poisson-Boltzmann’s
method or by the generalized Born model that account for the polar component of
the solvation free energy. The non-polar component of the solvation free energy
is estimated using the non-polar surface area of the complex, receptor, and ligand.
MM-PB/SA and MM-GB/SA energies employ either the single-trajectory or the
3-trajectory approach. In the single-trajectory method, conformational samples are
collected from the MD simulations of the complex alone, from which the receptor and
ligand components are separated during the energy calculations. In the 3-trajectory
approach, separate MD simulations are performed for the complex, the receptor, and
the ligand. Irrespective of the approach, the following Eq. 6.3 is used to compute the
binding free energy (∆Gbind ).
< > (< > < >)
∆G bind = ∆G complex − ∆G protein + ∆G ligand , (6.3)
< > < >
where ∆Gprotein and ∆Gligand are the total energies of the protein
< and >ligand and the
total free energy of the protein–ligand complex is given by ∆G complex . The angular
bracket <> indicates the energy is calculated from the structural ensemble derived
from MD simulations.
MD simulations, as mentioned earlier, can also be used to understand the binding
mechanism of the ligand to the receptor. Using unbiased MD simulations with an
explicit solvent model, Huang et al. [53] have made an effort to understand the path-
ways by which the ligands approach the protein active site and the conformational
changes that occur in protein during binding. The ligands xk263 (Fig. 6.6) and riton-
avir (Fig. 6.2) in complex with HIV protease were used. The studies reveal that the
two ligands bind to the protein by different mechanisms. The xk263 binds fast to
the protein with a semi-open flap conformation, indicating the induced-fit mecha-
nism, whereas ritonavir binds slowly to the protein in the open conformation via a
conformation selection mechanism. The HIV protease conformational changes and
the binding routes of the ligand (xk263) were sampled by Miao et al. [54] using the
Gaussian accelerated molecular dynamics (GaMD) method. The HIV protease struc-
ture is a homodimer with two loops or ‘flaps’. The three main flap conformations—
‘open’, ‘semi-open’, and ‘closed’ were found in this study. The apo-protein exhibits
the ‘semi-open’ conformation, whereas the holo-protein predominantly adopts the
‘closed’ conformation. A number of crucial intermediate states during the ligand
binding process were also discovered. The whole pathway of the ligand xk263 which
is a fast and tight binder of the HIV protease were successfully reproduced. The
GaMD simulation is a useful guide to further probe drug-receptor binding.
The entry of HIV into the T-cell is initiated by the interaction of the viral envelope
protein gp120 with the cell surface receptor CD4 as well as the co-receptors CCR5
or CXCR4. R5 virions and X4 virions, respectively, are the names given to viruses
that require CCR5 or CXCR4 for entry. According to reports [57], the dendrimer-
SPL7013, a microbicide binds to areas on the interface of the gp120-CD4 complex,
6 Role of Computational Modelling in Drug Discovery for HIV 171

Fig. 6.6 Structures of the ligands xk263 and HPCAR28

blocking viral entry into target cells. Fully atomistic molecular dynamics simula-
tions were employed by Nandy et al. [57] to evaluate the kinetics of dissociation and
energetics of the gp120-CD4 complex in the absence and presence of the dendrimer.
Molecular docking with steered and fully atomistic MD simulations was able to
predict that the dendrimer-SPL7013 does not bind to gp120 alone but binds strongly
to the R5 gp120 in the gp120-CD4 complex. This weakens the gp120-CD4 complex
and causes its dissociation. As a result, the gp120-CD4 complexes are not formed in
adequate number, to form across a virus-cell pair, thereby preventing viral entry. The
identification of the contact residues between CD4 and gp120 that were changed by
the binding of SPL7013’s binding to R5 gp120 were made possible by the atomistic
resolution offered by the MD simulations. When the binding energy was decom-
posed into its component elements, it became clear that the electrostatic component
makes the largest contribution to the total binding energy. The study thus provided a
mechanism of how SPL7013 prevents R5 HIV-1 from infecting target cells.
A common strategy in drug discovery is to perform an MD simulation of the
receptor-ligand complex to understand the binding affinity. Chen et al. [55] designed
a novel series of 52 dihydrofuran [3,4-d]pyrimidine (DHPY) analogues as NNRTIs.
A systematic in silico study comprising of 3D-QSAR, molecular docking, virtual
screening followed by filtering of the top ligands, and lastly MD simulations of the
complex were performed. They identified nine lead compounds using this compu-
tational strategy. Sirous et al. [58] using the structure-based combinatorial library
design method optimized a series of 3-hydroxypyran-4-one derivatives as HIV inte-
grase inhibitors. The method allowed the coupling of the combinatorial library design
with the quantum polarized ligand docking (QPLD) and MD simulations. HPCAR28
(Fig. 6.6) was identified as a potential lead in this experiment with an IC50 of
0.065 μM. A small library of 93 molecules having the 3-hydroxypyrimidine-2,4-
dione core targeting HIV-1 RT associated RNase-H was curated by Gao et al. [40].
To verify the accuracy of the docking methodology, these molecules were docked,
and MD simulations were run. The CoMSIA and CoMFA methodologies were used
to create 3D-QSAR models. Six new molecules were identified as leads in this study.
Wang et al. [52] reported in silico efforts with 38 N1-aryl-benzimidazoles as NNRTIs.
The protocol followed in this case was similar to Gao et al.; a 3D-QSAR model
172 A. Gomatam et al.

was built followed by a pharmacophore model to identify the structural features

related to the activity. This study identified positions on the aryl/benzimidazole ring
where appropriate substituents will enhance the inhibitory efficacy of the molecules.
These are hydrophobic groups at the linker of the C2-position of the benzimidazole
moiety; negatively charged and/or hydrogen-bond acceptor groups at the C6-position
of the benzimidazole moiety; small, positively charged, and/or hydrogen-bond donor
groups at the C2-position of the arylacetamide moiety.
A reliable pharmacophore model was created by Cele et al. [59] using MD
simulation ensembles and per-residue energy decomposition. The amino acids that
contribute to free energy of binding were the basis for the creation of the phar-
macophore library. A pharmacophoric screen for possible reverse transcriptase
was also conducted. To verify the system’s stability, docking and MD modelling
were applied to the complex of GSK952 with the protein. Utilizing the recognized
HIV-reverse transcriptase inhibitory activity, the technique was validated. Two hits
(ZINC46849657 and ZINC54359621) demonstrated a considerable potential to be
further investigated based on the binding free energy.
MD simulations may also be utilized to study the wild type and resistant or
different subtypes of the proteins to capture significant ligand interactions that could
subsequently be used to design a drug effective against the wild type or resistant
species. Halder et al. [60] have reported a study with the FDA approved protease
inhibitors—atazanavir, darunavir, and ritonavir. These were analysed for their activity
on other HIV protease subtypes like the South African subtype C (C-SA) and B
complexes using MD simulations. They have pointed out the specific affinities in
the subtypes using PCA, per-residue decomposition analysis, and hydrogen analysis
methods. The analysis identified major factors that contribute to increased binding
affinity of the compounds against C-SA protease over the B complex. These are
stable interactions with catalytic amino acid residues; increased electrostatic inter-
actions, and stable hydrogen-bond formation capacities with amino acid residues in
the binding cavity; stability in the flap movements caused by inhibitor binding and
decreased entropic cost.

6.5 Quantitative Structure–Activity Relationships (QSARs)

The QSAR formalism attempts to derive a mathematical relationship between the

structure of chemicals and their physiological behaviour in biological systems, such
as biological activity, disposition, and toxicity. Mathematically, the basis of QSAR is
a representation of the biological response of a chemical as a function of its chemical
attributes. The linear form of a QSAR equation (model) is given as

y = m 0 + m 1 x1 + m 2 x2 + m 3 x3 . . . m n xn , (6.4)
6 Role of Computational Modelling in Drug Discovery for HIV 173

Fig. 6.7 An illustration of the QSAR workflow

where y is the response being modelled and x1 , x2 , x3 ...xn are numerical representa-
tions of the structural features (also known as descriptors), and m 1 , m 2 , m 3 . . . m n are
the contributions (weights) of individual descriptors to the response. Once a model
of sufficient quality has been developed; it can be tested on hitherto untested or new
chemical entities [61]. Oftentimes, QSAR models can also direct lead design and
optimization by suggesting structural modifications to obtain the desired pharma-
cological activity, especially when the model has been developed on a congeneric
series of structurally related compounds. The correlation between the response and
structural features can be established using various chemometric methods [62]. The
original QSAR methodology pioneered by Hansch and Fujita used linear equations
to derive correlations between closely related compounds [63]. QSAR models are
often tested for their predictive ability on an external set of compounds. If a suit-
able external set is unavailable, the data is divided into a training set (used for model
building) and a test set (for model validation). Prior to deployment on unknown chem-
icals, any QSAR model must be rigorously validated for reliability, robustness, and
predictive ability using suitable validation metrics [61]. We provide an illustration
of the QSAR workflow in Fig. 6.7.

6.6 Pharmacophore Modelling

The concept of a pharmacophore was introduced by Paul Ehrlich, who defined it as

‘a molecular framework that carries the essential features (phoros) responsible for
a drug’s biological activity (pharmacon)’ [64]. Since then, the International Union
174 A. Gomatam et al.

of Pure and Applied Chemistry has redefined a pharmacophore as ‘the ensemble of

steric and electronic features that is necessary to ensure the optimal supramolecular
interactions with a specific biological target and to trigger (or block) its biological
response’ [65]. It is noteworthy that a pharmacophore does not refer to particular
functional groups, but rather the pattern of features in a molecule such as pres-
ence of hydrogen-bond donors or acceptors, hydrophobic (aromatic or aliphatic),
cationic, and anionic groups. A pharmacophore model consists of these features
(represented as spheres) in a specific three-dimensional pattern and can be used to
query test compounds or for screening a library of molecules (pharmacophore-based
virtual screening). A molecule that fits the spheres representing the pharmacophore
is considered a hit [66]. Depending on the available information regarding the target
of interest, a pharmacophore model can either be structure-based or ligand-based.
A structure-based pharmacophore model requires prior knowledge of the protein
target and comprises of features that best describe the major interactions between
the target and its ligands. If the structure of the target is unknown but active molecules
are known, a ligand-based pharmacophore model may be developed by mapping the
key pharmacophoric features of the active molecule. We outline the key steps in
pharmacophore modelling in Fig. 6.8.
There are numerous reports that use QSAR-based methods or a pharmacophore
modelling approach to design potent anti-HIV compounds. We summarize some of
the recent research in this area in Table 6.4.

Fig. 6.8 General workflow of pharmacophore modelling

Table 6.4 Some recently reported QSAR and pharmacophore modelling-based studies in drug discovery for HIV
Computational method Class Target Summary References
2D-QSAR 1238 diverse compounds HIV protease Models were built on 1238 [67]
protease inhibitors using
MLR, SVM, RF, and DNN
algorithms, with the DNN
2
model returning rtrain of 0.88
2 of 0.79
and rtest
2D-QSAR 346 diverse compounds HIV protease subtype B The best model was obtained [68]
mutant using XGB consensus
prediction, and they postulate
that SMILES attributes and
graph-based descriptors
sufficiently cover the
chemical space of the dataset
HQSAR, CoMFA and CoMSIA 82 tetrahydroimidazo(4,5,1-jk][1,4] HIV-1 reverse transcriptase The most robust model was [69]
benzodiazepinone derivatives obtained using HQSAR with
q 2 of 0.641
Molecular docking-based-QSAR 73 diarylpyrimidines HIV-1 reverse transcriptase The ANN model built with [70]
6 Role of Computational Modelling in Drug Discovery for HIV

six input layers, four hidden

layers, and one output layer
2 of
has MSE of 0.16 and rtest
0.89
(continued)
175
Table 6.4 (continued)
176

Computational method Class Target Summary References

CoMFA and CoMSIA 38 HIV-1 reverse transcriptase The CoMFA model returned [71]
S-dihydro-alkoxybenzyloxopyrimidines q 2 of 0.766 and r 2 0.949 for
the CoMFA model, while the
CoMSIA model returned q 2
of 0.827 and r 2 0.974
Combined Topomer CoMFA and 37 pyrrole derivatives HIV fusion inhibitors The final model has a high [72]
HQSARa 2
rtrain 2 of 0.96
of 0.96 and rtest
Topomer CoMFA, CoMSIA and 58 diarylpyrimidines Non-nucleoside reverse 2
The CoMSIA model has rtrain [73]
pharmacophore modelling transcriptase 2 of 0.73. The
of 0.95 and rtest
pharmacophore hypothesis
shows that the left-wing
aromatic ring and central
pyrimidine on the
diarylpyrimidine moiety are
essential for activity, and they
suggest modifications on the
right wing to improve activity
Pharmacophore modelling and 45 1,5-dihydrobenzo[b][1,4] HIV capsid protein The pharmacophore [74]
3D-QSAR diazepine-2,4-dione analogues hypothesis shows presence of
a hydrophobic site, two
aromatic rings and two
acceptor regions in all the
active molecules, and the
3D-QSAR model has rtrain 2 of
0.92
(continued)
A. Gomatam et al.
Table 6.4 (continued)
Computational method Class Target Summary References
Pharmacophore modelling, 66 metronidazole derivatives Non-nucleoside reverse A combined structure and [75]
atom-based QSAR, molecular transcriptase ligand-based strategy has
docking, MMGBSA scoring been followed, and the results
suggest that metronidazole
derivatives may be promising
NNRTIs
Pharmacophore modelling 10 sulfonamide derivatives HIV-1 glycoprotein 120 The designed molecules were [76]
synthesized using a green
approach, and the best
compound shows tenfold
increase in activity compared
to the standard
Combined molecular docking 27 diverse compounds identified as hits Chemokine receptor 5 and From the 27 tested [77]
and pharmacophore approach chemokine receptor 4 compounds, three were
identified as inhibitors with
IC50 values ranging from
10.64 to 64.56 μM
QSAR, molecular docking and 36 triazolothienopyrimidine derivatives HIV-1 reverse transcriptase In addition to the models, [78]
6 Role of Computational Modelling in Drug Discovery for HIV

pharmacophore modelling several insights on the key

structural features of these
moieties that contribute to
anti-HIV activity have been
provided
a HQSAR—Hologram QSAR
177
178 A. Gomatam et al.

6.7 The Emergence of Machine Learning in Drug

Discovery for HIV

In the past decade, machine learning (ML) has transformed drug discovery and
development, with real-world applications ranging from virtual screening and de
novo design to reaction prediction and retrosynthesis [79]. The explosive rise of
ML is due to a combination of factors. Recent years have seen vast improvements
in computing capacity and chemometric techniques, leading to the emergence of
newer and more powerful ML methods. ML-based models have been further aided
by the development of general-purpose statistical packages like the R [80] and Python
programming languages [81] which enable implementation of a wide range of ML
algorithms for classification and regression analyses [82]. Given the high attrition
rate in drug development, many pharmaceutical companies have begun to invest their
resources in leveraging the power of AI/ML to reduce development costs [83]. There
are several powerful AI-based tools for drug discovery and development today such
as AlphaFold (protein 3D structure prediction) [84], DeepChem (a python-based
tool for predictive modelling in drug discovery—https://github.com/deepchem/dee
pchem), and DeepTox (toxicity prediction using deep learning—http://www.bioinf.
jku.at/research/DeepTox/) [85].
Broadly speaking, ML is the practice of using algorithms to process data, learn,
and make a prediction about a desired outcome [83]. ML algorithms may be super-
vised or unsupervised, based on whether labels are assigned to the training data.
It is important to bear in mind that the quality of the model relies largely on the
quality of the input data and imposes an upper limit on the accuracy and general-
izability of any subsequently developed ML model. Therefore, the raw data must
be screened for errors, omissions, missing values, and data type conversions. Once
cleaned, the chemical data is represented as input for ML, preferably with open-
source implementations such as RDKit (https://www.rdkit.org) or Dscribe (https://
singroup.github.io/dscribe), and an appropriate ML algorithm is chosen [86]. There
is a plethora of ML algorithms to choose from, ranging from simple linear methods to
complex nonlinear neural net architectures. Although their black-box nature often-
times renders mechanistic interpretation unfeasible, nonlinear ML algorithms have
proven to be effective at delineating complicated relationships between biological
phenomena and molecular structure [83]. To build a robust ML model, it is essential
to avoid both overfitting and underfitting. The model parameters and hyperparam-
eters in the algorithm of choice (for instance, weights, and activation functions in
neural nets) must be tuned to their optimum values. Models are built on the training
set, and once a set of satisfactory models have been obtained, they are validated on
a test set of compounds not part of the training set. Once the models are finalized, it
is considered good practice to make the data and code publicly available for ease of
reproducibility [86]. We summarize the background of some popular ML algorithms
in the following section.
6 Role of Computational Modelling in Drug Discovery for HIV 179

Fig. 6.9 Best fit line

obtained using MLR

6.7.1 Multiple Linear Regression

Multiple linear regression (MLR) is one of the most used algorithms owing to its
simplicity, reproducibility, and ability to produce models that allow easy interpreta-
tion. MLR is an extension of simple linear regression to more than one dimension
and attempts to build a linear relationship between the dependent variable (target
property) and independent variables (molecular feature space) by fitting the data to
a straight line (shown in Fig. 6.9).
The best fit line is calculated using the slope intercept form and is given as follows:

yi = m 1 x1 + m 2 x2 + m 3 x3 . . . m n xn , (6.5)

where yi is the target property, m 1 , m 2 , m 3 ...m n are the regression coefficients and
x1 , x2 , x3 ...xn are the descriptors (features or independent variables). The goal of
MLR is to find the values of the coefficients in the MLR equation that minimize
the mean-squared-error (average of the squared error between the observed target
values and the values predicted by the model). In addition to assuming a linear
relationship between the dependent and independent variables, MLR also assumes
that the variables are not correlated (i.e. they do not show multicollinearity) [61, 62].

6.7.2 Logistic Regression

Logistic regression (LR) is a supervised ML algorithm that, despite its name, is used
for classification problems. LR is a transformation of linear regression in that it uses
a logistic or sigmoid function to model a binary output variable, which restricts the
value of y from 0 to 1, as shown in Eq. 6.6.
180 A. Gomatam et al.

Fig. 6.10 Graphs for linear and logistic regression

1
Sigmoid function[F(x)] = , (6.6)
1+ e−(β0 +β1 x)

where β0 + β1 x is analogous to y = mx + c for linear regression

In its most basic form, logistic regression is used to predict binary outcomes, but
it can also be extended to multiclass labels (multinomial logistic regression). Since
LR takes a linear combination of features and applies a nonlinear sigmoidal function,
it does not require a linear relationship between the target and predictor variables
(shown in Fig. 6.10).
The vertical axis denotes the probability of a given classification, and the hori-
zontal axis contains different values of x. Model predictions are interpreted as the
probability of a sample belonging to a particular class. LR is less prone to overfitting,
and its interpretability is a major advantage over black-box methods such as neural
networks, however, their simplistic nature may be a drawback when working with
rich and complex data [87, 88].

6.7.3 Naïve Bayes

The Naïve-Bayes is a probabilistic algorithm used primarily for classification tasks

and assigns the most likely class for each sample by applying the Bayes’ theorem.
The Bayes’ theorem is a formula for calculating conditional probabilities and is given
in Eq. 6.7.

P(B/A)P(A)
P(A/B) = , (6.7)
P(B)

where P( A/B) is posterior probability (probability of hypothesis A given B is an

observed outcome), P(B/ A) is likelihood probability (probability of event B given A
6 Role of Computational Modelling in Drug Discovery for HIV 181

is an observed outcome), P(A) is prior probability (probability of hypothesis before

observing evidence) and P(B) is marginal probability (probability of event B).
The Naïve-Bayes algorithm assumes that all the variables in the feature space
are independent of each other (hence the term ‘naïve’), an assumption which may
not always be true. Despite this, Naïve-Bayes performs well on datasets with non-
independent predictors and works well with small and/or noisy datasets. Naïve-
Bayes performs especially well when the input data is categorical. It is not the
algorithm of choice for high-dimensionality problems or when the feature space
comprises of many continuous variables, since the latter case requires mathematical
transformations on the input data [90].

6.7.4 Support Vector Machines

Support vector machine (SVM) is a supervised ML algorithm that is commonly used

for both classification and regression problems. SVM works by plotting each data
point in n-dimensional space where n is the number of features in the input data, the
numerical value of each feature is a particular coordinate, and the support vectors
are the coordinates of each observation [61]. The process of training an SVM model
is to identify a ‘hyperplane’ that maximizes the distance between the support vectors
of the two class labels [91] (illustrated in Fig. 6.11).
For data that is not linearly separable, SVM introduces a method known as the
kernel trick. The input data is mapped onto a higher dimensional space using a kernel
function and separated in that space using a maximum margin hyperplane. The most
used kernels include the linear kernel, the polynomial kernel, the radial basis function
kernel, and the sigmoid kernel. SVMs are one of the most widely used ML algorithms

Fig. 6.11 SVM illustration showing the hyperplane that best separates the two classes by
maximizing the distance between the support vectors
182 A. Gomatam et al.

owing to their effectiveness in high dimensions but are computationally expensive

and may not be the method of choice for large datasets [89].

6.7.5 Tree-Based Methods

Decision trees, useful for both classification and regression problems are intuitive
ML tools that utilize a tree-like structure for decision making. Analogous to trees in
nature, decision trees comprise of three types of nodes: a root node (from which the
tree starts), decision nodes (branches) that split the data into subsets, and terminal
nodes (the leaves) to assign the data to a target property (shown in Fig. 6.12) [61].
The algorithm starts by searching the feature space and selecting the feature that
best separates the classes, and assigns the identified feature to the root node. The
subsets created are searched again for identifying features for further separating the
data, these features are assigned to the decision nodes [92]. This process is repeated
iteratively until all the samples are predicted satisfactorily or if further partitioning
does not lead to an improved outcome [61].
Tree-based methods use goodness functions such as Gini scores, gain ratios, and
information gain [92]. Various algorithms can be used for constructing decision
trees, and among these, the most used is the random forest (RF) algorithm. RF is an
ensemble learning method wherein several decision trees (collectively known as a
forest) are built using bootstrapped samples of the data and a consensus prediction
is made. This solves the problem of overfitting while improving accuracy. RF is
commonly used owing to its superior performance and ease of implementation, as
there are only two parameters the user needs to define while building the forest: the
number of trees and the number of features in each tree [61].

Fig. 6.12 A typical decision tree

6 Role of Computational Modelling in Drug Discovery for HIV 183

6.7.6 Artificial Neural Networks

The idea of artificial neural networks (ANNs) originates from the functioning of the
neurons in an animal brain. The architecture of an ANN comprises of three essential
layers: the input layer, the hidden layers (may be one or more depending on the
complexity of the ANN), and the output layer. The input layer is the first layer in an
ANN and receives the training data. The hidden layers perform computations on the
input data and recognize patterns and are displayed by the output layer as the results
(shown in Fig. 6.13).
Each node in the input layer corresponds to an independent variable and is
connected to the hidden layer, and each node in the hidden layer denotes a dependent
variable and is connected to the output layer. Each neuron in the neural network
has a ‘weight’ parameter associated with it and receive inputs as signals in accor-
dance with their respective weights. A summation function is used to calculate the
combined input signals which are passed through an activation function. The activa-
tion function then maps the input to produce an output from the neuron. Examples of
activation functions include the hyperbolic tangent function, the softmax function,
and the rectified linear unit function. Models learn by adjusting the weights of the
neurons, which result in modified outputs for each input. A schematic representation
of an artificial neuron is illustrated in Fig. 6.14.
Despite tremendous increase in the use of ANNs in ML tasks, some of its short-
comings remain unsolved. ANNs are referred to as ‘black-boxes’, because the
input and output from the neuron are known, but not what happens inside it. As
a result, model interpretation for chemistry problems can oftentimes be challenging
[61, 89, 93–95].
There are several reports of ML methods used to tackle the resistance problem
in HIV. This chapter will cover some of the more recent work in this area. For a
description of previous works, we refer readers to a review by Reimenschneider and
Heider [96] (summarized in Table 6.5).

Fig. 6.13 Basic neural network architecture

184 A. Gomatam et al.

Fig. 6.14 Schematic representation of an artificial neuron

Table 6.5 Summary of reported literature on ML methods

Data Target/s ML algorithm Summary References
55,000 reverse HIV-1 reverse NB, LR, and An attempt has been [97]
transcriptase transcriptase RF made to differentiate
sequences between RTI
experienced and RTI
naïve population,
along with the
discovery of six new
mutations associated
with drug resistance
Genotype–phenotype HIV-1 reverse RF and SVM Drug fold values were [98]
data for 21 drugs transcriptase, modelled as a function
obtained from HIV protease, of protein sequence
StanfordDB HIV integrase represented as
physicochemical
properties using a
weighted ML
approach. The model
was able to build
satisfactorily
predictive models for
13 out 21 HIV
approved drugs
(continued)
6 Role of Computational Modelling in Drug Discovery for HIV 185

Table 6.5 (continued)

Data Target/s ML algorithm Summary References
NIAID ChemDB HIV, HIV-1 NB, DT, RF, Models for HIV [99]
Opportunistic wild-type SVM, kNNa , inhibition were
infection and cell-based and DNNa , developed using
tuberculosis reverse consensus publicly available data,
Therapeutics database transcriptase and a comparison of
DNA different ML methods
polymerase demonstrated that
inhibition SVM, deep learning,
and consensus models
showed the most
promising results
Stanford HIV drug HIV-1 reverse RF and SVM Sequences from HIV [100]
resistance database transcriptase, isolates and their
HIV protease, susceptibility to
HIV integrase antiretroviral drugs
were modelled using
weighted categorical
kernel functions,
which resulted in
superior models for
predicting drug
resistance, especially
in the case of HIV-1
RT
Stanford HIV Drug HIV-1 reverse RF Amino acid sequences [101]
Resistance Database transcriptase were represented as
and HIV short fragments and
protease used as descriptors to
model resistance of RT
and protease to
marketed drugs. They
conclude that model
performance is more
sensitive to certain
drugs than descriptor
type
Stanford HIV Drug HIV-1 reverse MLP, BRNNa CNNs were reported [102]
Resistance Database transcriptase and CNN as the best performing
and HIV DL method. The
protease black-box problem
was addressed by
feature importance
analysis for
identification of
known and novel
mutations as
biologically relevant
features, giving an
interpretable DL
model
(continued)
186 A. Gomatam et al.

Table 6.5 (continued)

Data Target/s ML algorithm Summary References
HIV-1 RT mutant HIV-1 reverse NB Loss of HIV-RT [103]
susceptibility data transcriptase activity on mutations
from ChEMBL in three major
residues: Y181, K103,
and L100 was studied.
The models had an
average ROC AUC of
0.920
Darunavir-bound HIV protease LR Data collected from [104]
HIV-1 protease atomistic simulations
variants of HIV protease
variants in complex
with darunavir were
used as input variables
in ML, and
mechanistic insignts
were provided on how
alterations in the
darunavir-protease
complex can affect
drug binding
Darunavir-bound HIV protease Ordinary least ML was coupled with [105]
HIV-1 protease squares parallel MD
variants simulations, and the
linear regression
model correlating
non-covalent
drug-receptor
interactions of
darunavir for HIV
protease and its
mutants was accurate
and performed well on
a test set of protease
variants
Resistance data for six HIV-1 reverse RF A web server named [106]
antiretroviral drug transcriptase, SHIVA was developed
classes comprising of HIV protease, for resistance
23 drugs HIV integrase prediction of some
commonly used
antiretroviral drugs.
SHIVA was found to
be superior to other
popular server-based
prediction tools such
as geno2pheno,
HIVdb, and
WebPSSM
(continued)
6 Role of Computational Modelling in Drug Discovery for HIV 187

Table 6.5 (continued)

Data Target/s ML algorithm Summary References
Protein sequences HIV-1 reverse Binary The cross-resistance [107]
with data for ten drugs transcriptase, relevance problem in HIV was
were collected from HIV protease classifiers, addressed by building
the HIV drug classifier multi-label
resistance database chains, and classification models.
ensembles of Multi-label learning
classifier was able to improve
chains classification accuracy
as compared to binary
classifiers
Stanford HIV drug HIV protease PLSa , RF, A model for resistance [108]
resistance database LGBMa , and prediction was
SVR developed based on a
homology modelling
and molecular field
mapping approach.
The model based on
the CoMFA
methodology was
robust, with the
LGBM algorithm
performing the best
Stanford HIV Drug HIV-1 reverse ANN A subtype-specific [109]
Resistance database transcriptase, approach for the
HIV protease prediction of fold
resistance was
followed, with the
ANN model
comparable to reported
models in the literature
a kNN k-nearest neighbour, DNN deep neural network, BRNN bidirectional recurrent neural

networks, PLS partial least squares, LGBM light gradient boosting machine
188 A. Gomatam et al.

6.8 Conclusion

The surge in computational power along with the abundance of available biological
data has paved the way for in silico methods to play an instrumental role in anti-HIV
drug discovery and resistance mapping. Remarkable results have been achieved in
our understanding of HIV biology, the impact of mutations, and subsequent devel-
opment of resistance and in the design of novel molecules that are active against
these resistant strains of the virus. The identification and elucidation of new drug
targets has greatly benefitted structure-based methods such as molecular docking and
MD simulations, whereas QSAR modelling and other ML-based methods have been
aided by the development of publicly available databases such as the StanfordDB.
However, several challenges remain. Data quality is an ongoing concern, and any
theoretical model must be validated rigorously and evaluated for their usefulness
in a real-life scenario. In this regard, it is essential that a collaborative framework
is established that enables rapid experimental validation of hypotheses generated in
silico. Hopefully, computational approaches will continue to play an important role
and will enable improved treatment of HIV and eventually, its eradication.

References

1. Charneau P, Borman AM, Quillent C, Guétard D, Chamaret S, Cohen J, Rémy G, Montagnier

L, Clavel F (1994) Isolation and envelope sequence of a highly divergent HIV-1 isolate:
definition of a new HIV-1 group. Virology 205(1):247–253. https://doi.org/10.1006/viro.1994.
1640
2. HIV/AIDS. https://www.who.int/news-room/fact-sheets/detail/hiv-aids
3. Seitz R (2016) Human immunodeficiency virus (HIV). Transfus Med Hemotherapy
43(3):203–222. https://doi.org/10.1159/000445852
4. Waymack J, Sundareshan V (2021) Acquired immune deficiency syndrome; StatPearls
Publishing
5. How Is HIV Transmitted? https://www.hiv.gov/hiv-basics/overview/about-hiv-and-aids/how-
is-hiv-transmitted
6. Rossi E, Meuser ME, Cunanan CJ, Cocklin S (2021) Structure, function, and interactions of
the Hiv-1 capsid protein. Life 11(2):1–25. https://doi.org/10.3390/life11020100
7. Kirchhoff F (2016) Encyclopedia of AIDS. Encycl AIDS 2016 (January). https://doi.org/10.
1007/978-1-4614-9610-6
8. Ugolini S, Mondor I, Sattentau QJ (1999) HIV-1 attachment : another look 99:144–149.
https://doi.org/10.1016/S0966-842X(99)01474-2
9. HIV/AIDS Glossary. https://clinicalinfo.hiv.gov/en/glossary/life-cycle
10. De Clercq E (2009) Anti-HIV drugs: 25 compounds approved within 25 years after the
discovery of HIV. Int J Antimicrob Agents 33(4):307–320. https://doi.org/10.1016/j.ijanti
micag.2008.10.010
11. Cilento ME, Kirby KA, Sarafianos SG (2021) Avoiding drug resistance in HIV reverse
transcriptase. Chem Rev 121(6):3271–3296. https://doi.org/10.1021/acs.chemrev.0c00967
12. Portegies P (2002) Antiretroviral therapeutics. J Neurovirol 8(SUPPL. 2):148–150. https://
doi.org/10.1080/13550280290167966
13. Gu SX, Zhu YY, Wang C, Wang HF, Liu GY, Cao S, Huang L (2020) Recent discoveries in
HIV-1 reverse transcriptase inhibitors. Curr Opin Pharmacol 54:166–172. https://doi.org/10.
1016/j.coph.2020.09.017
6 Role of Computational Modelling in Drug Discovery for HIV 189

14. Maldarelli F (2006) HIV drug resistance. Handb Pediatr HIV Care, 2nd ed, pp 397–414.
https://doi.org/10.1017/CBO9780511544781.016
15. Preston BD, Poiesz BJ, Loeb LA (1998) Fidelity of HIV-1 reverse transcriptase. Science
(80-.) 242(4882):1168–1171. https://doi.org/10.1126/science.2460924
16. Vandamme AM, Van Laethem K, De Clercq E (1999) Managing resistance to anti-HIV drugs:
an important consideration for effective disease management. Drugs 57(3):337–361. https://
doi.org/10.2165/00003495-199957030-00006
17. Collier DA, Monit C, Gupta RK (2019) The impact of HIV-1 drug escape on the global
treatment landscape. Cell Host Microbe 26(1):48–60. https://doi.org/10.1016/j.chom.2019.
06.010
18. Anderson A (2003) The process of structure- based drug design. Chem Biol 10:787–797.
https://doi.org/10.1016/j.chembiol.2003.09.002
19. Koshland DE (1995) The key-lock theory and the induced fit theory. Angew Chemie Int Ed
English 33(23–24):2375–2378. https://doi.org/10.1002/anie.199423751
20. Saikia S, Bordoloi M (2019) Molecular docking: challenges, advances and its use in drug
discovery perspective. Curr Drug Targets 20(5):501–521. https://doi.org/10.2174/138945011
9666181022153016
21. Śledź P, Caflisch A (2018) Protein structure-based drug design: from docking to molecular
dynamics. Curr Opin Struct Biol 48:93–102. https://doi.org/10.1016/j.sbi.2017.10.010
22. Karplus M, McCammon JA (2010) Molecular dynamics simulations of biomolecules. Mol
Simul 36(13):1035–1044. https://doi.org/10.1080/08927022.2010.501797
23. Talele T, Khedkar S, Rigby A (2010) Successful applications of computer aided drug
discovery: moving drugs from concept to the clinic. Curr Top Med Chem 10(1):127–141.
https://doi.org/10.2174/156802610790232251
24. Fan J, Fu A, Zhang L (2019) Progress in molecular docking. Quant Biol 7(2):83–89. https://
doi.org/10.1007/s40484-019-0172-y
25. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, Ferrin TE (2004)
UCSF chimera—a visualization system for exploratory research and analysis. J Comput Chem
25(13):1605–1612. https://doi.org/10.1002/jcc.20084
26. Almerico AM, Tutone M, Lauria A (2008) Docking and multivariate methods to explore HIV-
1 drug-resistance: a comparative analysis. J Comput Aided Mol Des 22(5):287–297. https://
doi.org/10.1007/s10822-008-9186-7
27. Morris GM, Huey R, Lindstrom W, Sanner MF, Belew R, Goodsell D, Olson A (2009)
AutoDock4 and AutoDockTools4: automated docking with selective receptor flexibility. J
Comput Chem. https://doi.org/10.1002/jcc.21256
28. Sybyl-X Molecular Modeling Software Packages. TRIPOS Associates, Inc.
29. Trott O, Olson AJ (2009) Software news and update AutoDock vina: improving the speed and
accuracy of docking with a new scoring function, efficient optimization, and multithreading.
J Comput Chem 31(2). https://doi.org/10.1002/jcc.21334
30. Bitencourt-Ferreira G, Filgueira de Azevedo Jr W (2019) How docking programs work. In:
Docking screens for drug discovery, Springer, pp 35–50
31. Forli SR (2010) AutoDock VS: an automated tool for preparing autodock virtual screenings
32. Friesner RA, Banks JL, Murphy RB, Halgren TA, Klicic JJ, Mainz DT, Repasky MP, Knoll
EH, Shelley M, Perry JK, Shaw DE, Francis P, Shenkin PS (2004) Glide: a new approach for
rapid, accurate docking and scoring. 1. method and assessment of docking accuracy. J Med
Chem 47(7):1739–1749. https://doi.org/10.1021/jm0306430
33. Vora J, Patel S, Sinha S, Sharma S, Srivastava A, Chhabria M, Shrivastava N (2019) Molecular
docking, QSAR and ADMET based mining of natural compounds against prime targets of
HIV. J Biomol Struct Dyn 37(1):131–146. https://doi.org/10.1080/07391102.2017.1420489
34. Tarasova O, Poroikov V, Veselovsky A (2018) Molecular docking studies of HIV-1 resistance
to reverse transcriptase inhibitors: mini-review. Molecules 23(5):11–13. https://doi.org/10.
3390/molecules23051233
35. Singh VK, Srivastava R, Gupta PSS, Naaz F, Chaurasia H, Mishra R, Rana MK, Singh RK
(2021) Anti-HIV potential of diarylpyrimidine derivatives as non-nucleoside reverse tran-
scriptase inhibitors: design, synthesis, docking, TOPKAT analysis and molecular dynamics
190 A. Gomatam et al.

simulations. J Biomol Struct Dyn 39(7):2430–2446. https://doi.org/10.1080/07391102.2020.

1748111
36. Fraì̈czek T, Siwek A, Paneth P (2013) Assessing molecular docking tools for relative biological
activity prediction: a case study of triazole HIV-1 NNRTIs. J Chem Inf Model 53(12):3326–
3342. https://doi.org/10.1021/ci400427a
37. Liu G, Wang W, Wan Y, Ju X, Gu S (2018) Application of 3D-QSAR, pharmacophore,
and molecular docking in the molecular design of diarylpyrimidine derivatives as HIV-1
nonnucleoside reverse transcriptase inhibitors. Int J Mol Sci 19(5). https://doi.org/10.3390/
ijms19051436
38. Makarasen A, Kuno M, Patnin S, Reukngam N, Khlaychan P, Deeyohe S, Intachote P,
Saimanee B, Sengsai S, Boonsri P, Chaivisuthangkura A, Sirithana W, Techasakul S (2019)
Molecular docking studies and synthesis of amino-oxy-diarylquinoline derivatives as potent
non-nucleoside HIV-1 reverse transcriptase inhibitors. Drug Res (Stuttg) 69(12):671–682.
https://doi.org/10.1055/a-0968-1150
39. Shanty AA, Raghu KG, Mohanan PV (2019) Synthesis, characterization: spectral and theo-
retical, molecular docking and in vitro studies of copper complexes with HIV RT enzyme. J
Mol Struct 1197:154–163. https://doi.org/10.1016/j.molstruc.2019.06.097
40. Gao Y, Chen Y, Tian Y, Zhao Y, Wu F, Luo X, Ju X, Liu G (2019) In Silico study of 3-
hydroxypyrimidine-2,4-diones as inhibitors of HIV RT-associated RNase H using molec-
ular docking, molecular dynamics, 3D-QSAR, and pharmacophore models. New J Chem
43(43):17004–17017. https://doi.org/10.1039/c9nj03353j
41. Faghihi K, Safakish M, Zebardast T, Hajimahdi Z, Zarghi A (2019) Molecular docking and
QSAR study of 2-benzoxazolinone, quinazoline and diazocoumarin derivatives as anti-HIV-1
agents. Iran J Pharm Res 18(3):1253–1263. https://doi.org/10.22037/ijpr.2019.1100746
42. Turkovic N, Ivkovic B, Kotur-Stevuljevic J, Tasic M, Marković B, Vujic Z (2020) Molecular
docking, synthesis and anti-HIV-1 protease activity of novel chalcones. Curr Pharm Des
26(8):802–814. https://doi.org/10.2174/1381612826666200203125557
43. Hajimahdi Z, Zabihollahi R, Aghasadehi MR, Zarghi A (2019) Design, synthesis, docking
studies and biological activities novel 2,3-Diaryl-4-quinazolinone derivatives as anti-HIV-1
agents. Curr HIV Res 17(3)
44. McCammon JA, Gelin BR, Karplus M (1977) Dynamics of folded proteins. Nature 267.
https://doi.org/10.1038/267585a0
45. Wang J, Wolf RM, Caldwell JW, Kollman PA, Case DA (2004) Development and testing of a
general amber force field. J Comput Chem 56531(9):1157–1174. https://doi.org/10.1002/jcc.
20035
46. Vanommeslaeghe K, Hatcher E, Acharya C, Kundu S, Zhong S, Shim J, Darian E, Guvench
O, Lopes P, Vorobyov I, Mackerell AD Jr (2010) CHARMM general force field: a force field
for drug-like molecules compatible with the CHARMM all-atom additive biological force
fields. J Comput Chem 31(4):671–690. https://doi.org/10.1002/jcc.21367
47. Schmid N, Eichenberger AP, Choutko A, Riniker S, Winger M, Mark AE, Van Gunsteren
WF (2011) Definition and testing of the GROMOS force-field versions 54A7 and 54B7. Eur
Biophys J 40(7):843–856. https://doi.org/10.1007/s00249-011-0700-9
48. Case DA, Cheatham TE, Darden T, Gohlke H, Luo R, Merz KM, Onufriev A, Simmerling C,
Wang B, Woods RJ (2005) The amber biomolecular simulation programs. J Comput Chem
26(16):1668–1688. https://doi.org/10.1002/jcc.20290
49. Phillips JC, Braun R, Wang W, Gumbart J, Tajkhorshid E, Villa E, Chipot C, Skeel RD,
Kalé L, Schulten K (2005) Scalable molecular dynamics with NAMD. J Comput Chem
26(16):1781–1802. https://doi.org/10.1002/jcc.20289
50. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan S, Karplus M (1983)
CHARMM: a program for macromolecular energy, minimization, and dynamics calculations.
J Comput Chem 4(2):187–217. https://doi.org/10.1002/jcc.540040211
51. Christen M, Hünenberger PH, Bakowies D, Baron R, Bürgi R, Geerke DP, Heinz TN,
Kastenholz MA, Kräutler V, Oostenbrink C, Peter C, Trzesniak D, Van Gunsteren WF
(2005) The GROMOS software for biomolecular simulation: GROMOS05. J Comput Chem
26(16):1719–1751. https://doi.org/10.1002/jcc.20303
6 Role of Computational Modelling in Drug Discovery for HIV 191

52. Wang W, Tian Y, Wan Y, Gu S, Ju X, Luo X, Liu G (2019) Insights into the key structural
features of N1-ary-benzimidazols as HIV-1 NNRTIs using molecular docking, molecular
dynamics, 3D-QSAR, and pharmacophore modeling. Struct Chem 30(1):385–397. https://
doi.org/10.1007/s11224-018-1204-3
53. Huang YMM, Raymundo MAV, Chen W, Chang CEA (2017) Mechanism of the associa-
tion pathways for a pair of fast and slow binding ligands of HIV-1 protease. Biochemistry
56(9):1311–1323. https://doi.org/10.1021/acs.biochem.6b01112
54. Miao Y, Huang YMM, Walker RC, McCammon JA, Chang CEA (2018) Ligand binding
pathways and conformational transitions of the HIV protease. Biochemistry 57(9):1533–1541.
https://doi.org/10.1021/acs.biochem.7b01248
55. Chen Y, Tian Y, Gao Y, Wu F, Luo X, Ju X, Liu G (2020) In silico design of novel HIV-1
NNRTIs based on combined modeling studies of Dihydrofuro[3,4-d]Pyrimidines. Front Chem
8(March):1–17. https://doi.org/10.3389/fchem.2020.00164
56. Martis EAF, Coutinho EC (2019) Free energy-based methods to understand drug resistance
mutations, 1–24. https://doi.org/10.1007/978-3-030-05282-9_1
57. Nandy B, Saurabh S, Sahoo AK, Dixit NM, Maiti PK (2015) The SPL7013 dendrimer desta-
bilizes the HIV-1 Gp120-CD4 complex. Nanoscale 7(44):18628–18641. https://doi.org/10.
1039/c5nr04632g
58. Sirous H, Chemi G, Gemma S, Butini S, Debyser Z, Christ F, Saghaie L, Brogi S, Fassihi A,
Campiani G, Brindisi M (2019) Identification of novel 3-hydroxy-pyran-4-one derivatives as
potent HIV-1 integrase inhibitors using in silico structure-based combinatorial library design
approach. Front Chem 7(August):1–20. https://doi.org/10.3389/fchem.2019.00574
59. Cele FN, Ramesh M, Soliman MES (2016) Per-residue energy decomposition pharmacophore
model to enhance virtual screening in drug discovery: a study for identification of reverse
transcriptase inhibitors as potential anti-hiv agents. Drug Des Devel Ther 10:1365–1377.
https://doi.org/10.2147/DDDT.S95533
60. Halder AK, Honarparvar B (2019) Molecular alteration in drug susceptibility against subtype
B and C-SA HIV-1 proteases: MD study. Struct Chem 30(5):1715–1727. https://doi.org/10.
1007/s11224-019-01305-0
61. Roy K, Kar S, Das RN (2015) Understanding the basics of QSAR for applications in
pharmaceutical sciences and risk assessment https://doi.org/10.1016/C2014-0-00286-9
62. Verma J, Khedkar V, Coutinho E (2010) 3D-QSAR in drug design—a review. Curr Top Med
Chem 10(1):95–115. https://doi.org/10.2174/156802610790232260
63. Hansch C, Fujita T (1964) ρ-σ-π analysis. a method for the correlation of biological activity
and chemical structure. J Am Chem Soc, 86(8):1616–1626. https://doi.org/10.1021/ja0106
2a035
64. Langer T, Hoffmann RD Pharmacophores and Pharmacophore Searches
65. Wermuth CG, Ganellin CR, Lindberg P, Mitscher LA (1998) Glossary of terms used in
medicinal chemistry. Pure Appl Chem 70(5):1129–1143
66. Qing X, Lee XY, De Raeymaeker J, Tame JR, Zhang KY, De Maeyer M, Voet AR (2014) Phar-
macophore modeling: advances, limitations, and current utility in drug discovery. J Receptor
Ligand Channel Res 7:81–92. https://doi.org/10.2147/JRLCR.S46843
67. Tian Y, Zhang S, Yin H, Yan A (2020) Quantitative structure-activity relationship (QSAR)
models and their applicability domain analysis on HIV-1 protease inhibitors by machine
learning methods. Chemom Intell Lab Syst 196:103888. https://doi.org/10.1016/j.chemolab.
2019.103888
68. Halder AK (2018) Finding the structural requirements of diverse HIV-1 protease inhibitors
using multiple QSAR modelling for lead identification. SAR QSAR Environ Res 29(11):911–
933. https://doi.org/10.1080/1062936X.2018.1529702
69. Tong J, Lei S, Qin S, Wang Y (2018) QSAR studies of TIBO derivatives as HIV-1 reverse
transcriptase inhibitors using HQSAR. CoMFA and CoMSIA J Mol Struct 1168:56–64.
https://doi.org/10.1016/j.molstruc.2018.05.005
70. Beglari M, Goudarzi N, Shahsavani D, Arab Chamjangali M, Dousti R (2020) QSAR modeling
of anti-HIV activity for DAPY-like derivatives using the mixture of ligand-receptor binding
192 A. Gomatam et al.

information and functional group features as a new class of descriptors. Netw Model Anal
Heal Inf Bioinforma 9(1). https://doi.org/10.1007/s13721-020-00261-8
71. Wang Y, Chang J, Wang J, Zhong P, Zhang Y, Lai CC, He Y (2018) 3D-QSAR studies of
S-DABO derivatives as non-nucleoside HIV-1 reverse transcriptase inhibitors. Lett Drug Des
Discov 16(8):868–881. https://doi.org/10.2174/1570180815666180810112321
72. Han D, Tan J, Zhou Z, Li C, Zhang X, Wang C (2018) Combined Topomer CoMFA and
hologram QSAR studies of a series of pyrrole derivatives as potential HIV fusion inhibitors.
Med Chem Res 27(7):1770–1781. https://doi.org/10.1007/s00044-018-2190-0
73. Liu G, Wan Y, Wang W, Fang S, Gu S, Ju X (2019) Docking-based 3D-QSAR and phar-
macophore studies on diarylpyrimidines as non-nucleoside inhibitors of HIV-1 reverse
transcriptase. Mol Divers 23(1):107–121. https://doi.org/10.1007/s11030-018-9860-1
74. Bhole RP, Bonde CG, Bonde SC, Chikhale RV, Wavhale RD (2021) Pharmacophore model
and atom-based 3D quantitative structure activity relationship (QSAR) of human immunod-
eficiency virus-1 (HIV-1) capsid assembly inhibitors. J Biomol Struct Dyn 39(2):718–727.
https://doi.org/10.1080/07391102.2020.1715258
75. Cutinho PF, Roy J, Anand A, Cheluvaraj R, Murahari M, Chimatapu HSV (2020) Design
of metronidazole derivatives and flavonoids as potential non-nucleoside reverse transcrip-
tase inhibitors using combined ligand- and structure-based approaches. J Biomol Struct Dyn
38(6):1626–1648. https://doi.org/10.1080/07391102.2019.1614094
76. Vangala R, Sivan SK, Peddi SR, Manga V (2020) Computational design, synthesis and eval-
uation of new sulphonamide derivatives targeting HIV-1 Gp120. J Comput Aided Mol Des
34(1):39–54. https://doi.org/10.1007/s10822-019-00258-0
77. Mirza MU, Saadabadi A, Vanmeert M, Salo-Ahen OMH, Abdullah I, Claes S, De Jonghe
S, Schols D, Ahmad S, Froeyen M (2020) Discovery of HIV entry inhibitors via a hybrid
CXCR4 and CCR5 receptor pharmacophore-based virtual screening approach. Eur J Pharm
Sci 155(July):105537. https://doi.org/10.1016/j.ejps.2020.105537
78. Ravichandran V, Rohini K, Harish R, Parasuraman S, Sureshkumar K (2019) Insights into
the key structural features of triazolothienopyrimidines as anti-HIV agents using QSAR,
molecular docking, and pharmacophore modeling. Struct Chem 30(4):1471–1484. https://
doi.org/10.1007/s11224-019-01304-1
79. Deng J, Yang Z, Ojima I, Samaras D, Wang F (2022) Artificial intelligence in drug discovery:
applications and techniques. Brief Bioinform 23(1):1–65. https://doi.org/10.1093/bib/bba
b430
80. R: A language and environment for statistical computing (2013)
81. Sanner MF (1999) Python : a programming language for software integration and develop-
ment. J Mol Graph Model 17(1):57–61
82. Dixon SL, Duan J, Smith E, Von Bargen CD, Sherman W, Repasky MP (2016) AutoQSAR: an
automated machine learning tool for best-practice quantitative structure-activity relationship
modeling. Future Med Chem 8(15):1825–1839. https://doi.org/10.4155/fmc-2016-0093
83. Vamathevan J, Clark D, Czodrowski P, Dunham I, Ferran E, Lee G, Li B, Madabhushi
A, Shah P, Spitzer M, Zhao S (2019) Applications of machine learning in drug discovery
and development. Nat Rev Drug Discov 18(6):463–477. https://doi.org/10.1038/s41573-019-
0024-5
84. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K,
Bates R, Žídek A, Potapenko A, Bridgland A, Meyer C, Kohl SAA, Ballard AJ, Cowie A,
Romera-Paredes B, Nikolov S, Jain R, Adler J, Back T, Petersen S, Reiman D, Clancy E,
Zielinski M, Steinegger M, Pacholska M, Berghammer T, Bodenstein S, Silver D, Vinyals O,
Senior AW, Kavukcuoglu K, Kohli P, Hassabis D (2021) Highly accurate protein structure
prediction with AlphaFold. Nature 596(7873):583–589. https://doi.org/10.1038/s41586-021-
03819-2
85. Mayr A, Klambauer G, Unterthiner T, Hochreiter S (2016) DeepTox: toxicity prediction using
deep learning. Front Environ Sci 3(FEB). https://doi.org/10.3389/fenvs.2015.00080
86. Artrith N, Butler KT, Coudert FX, Han S, Isayev O, Jain A, Walsh A (2021) Best practices in
machine learning for chemistry. Nat Chem 13(6):505–508. https://doi.org/10.1038/s41557-
021-00716-z
6 Role of Computational Modelling in Drug Discovery for HIV 193

87. Belyadi H, Haghighat A (2021) Machine learning guide for oil and gas using python; Gulf
Professional Publishing. https://doi.org/10.1016/c2019-0-03617-5
88. Subasi A (2020) Practical machine learning for data analysis using python. https://doi.org/
10.1016/B978-0-12-821379-7.00008-4
89. Carracedo-Reboredo P, Liñares-Blanco J, Rodríguez-Fernández N, Cedrón F, Novoa FJ,
Carballal A, Maojo V, Pazos A, Fernandez-Lozano C (2021) A review on machine learning
approaches and trends in drug discovery. Comput Struct Biotechnol J 19:4538–4558. https://
doi.org/10.1016/j.csbj.2021.08.011
90. Rhys HI (2020) Machine learning with R, the Tidyverse and MLR; Manning Publications
91. Pisner DA, Schnyer DM (2019) Support vector machine; Elsevier Inc., 2019. https://doi.org/
10.1016/B978-0-12-815739-8.00006-7
92. Djuris J, Ibric S, Djuric Z (2013) Neural computing in pharmaceutical products and process
development; Woodhead Publishing Limited https://doi.org/10.1533/9781908818324.91
93. Yacim JA, Boshoff DGB (2018) Impact of artificial neural networks training algorithms on
accurate prediction of property values. J Real Estate Res 40(3):375–418. https://doi.org/10.
1080/10835547.2018.12091505
94. Puri M, Pathak Y, Sutariya VK, Tipparaju S, Moreno W (2015) Artificial neural network for
drug design, delivery and disposition. https://doi.org/10.1016/C2014-0-00253-5
95. Krogh A (2008) What are artificial neural networks? Nat Biotechnol 26(2):195–197. https://
doi.org/10.1038/nbt1386
96. Riemenschneider M, Heider D (2016) Current approaches in computational drug resistance
prediction in HIV. Curr HIV Res 14(4):307–315
97. Blassel L, Tostevin A, Villabona-Arenas CJ, Peeters M, Hué S, Gascuel O (2021) Using
machine learning and big data to explore the drug resistance landscape in HIV. PLoS Comput
Biol 17(8):1–21. https://doi.org/10.1371/journal.pcbi.1008873
98. Cai Q, Yuan R, He J, Li M, Guo Y (2021) Predicting HIV drug resistance using weighted
machine learning method at target protein sequence-level. Mol Divers 25(3):1541–1551.
https://doi.org/10.1007/s11030-021-10262-y
99. Zorn KM, Lane TR, Russo DP, Clark AM, Makarov V, Ekins S (2019) Multiple machine
learning comparisons of HIV cell-based and reverse transcriptase data sets. Mol Pharm
16(4):1620–1632. https://doi.org/10.1021/acs.molpharmaceut.8b01297
100. Ramon E, Belanche-Muñoz L, Pérez-Enciso M (2019) HIV drug resistance prediction with
categorical kernel functions. BMC Bioinformatics 410(20):233–244. https://doi.org/10.1007/
978-3-030-17935-9_22
101. Tarasova O, Biziukova N, Filimonov D, Poroikov V (2018) A computational approach for
the prediction of HIV resistance based on amino acid and nucleotide descriptors. Molecules
23(11). https://doi.org/10.3390/molecules23112751
102. Steiner MC, Gibson KM (2020) Techniques on HIV-1 sequence data, pp 1–24
103. Kaiser TM, Burger PB, Butch CJ, Pelly SC, Liotta DC (2018) A machine learning approach for
predicting HIV reverse transcriptase mutation susceptibility of biologically active compounds.
J Chem Inf Model 58(8):1544–1552. https://doi.org/10.1021/acs.jcim.7b00475
104. Whitfield TW, Ragland DA, Zeldovich KB, Schiffer CA (2020) Characterizing protein-ligand
binding using atomistic simulation and machine learning: application to drug resistance in
HIV-1 protease. J Chem Theory Comput 16(2):1284–1299. https://doi.org/10.1021/acs.jctc.
9b00781
105. Leidner F, Kurt Yilmaz N, Schiffer CA (2021) Deciphering complex mechanisms of resistance
and loss of potency through coupled molecular dynamics and machine learning. J Chem
Theory Comput 17(4):2054–2064. https://doi.org/10.1021/acs.jctc.0c01244
106. Riemenschneider M, Hummel T, Heider D (2016) SHIVA—A web application for drug resis-
tance and tropism testing in HIV. BMC Bioinfo 17(1):1–6. https://doi.org/10.1186/s12859-
016-1179-2
107. Riemenschneider M, Senge R, Neumann U, Hüllermeier E, Heider D (2016) Exploiting HIV-1
protease and reverse transcriptase cross-resistance information for improved drug resistance
prediction by means of multi-label classification. BioData Min. 9(1):1–6. https://doi.org/10.
1186/s13040-016-0089-1
194 A. Gomatam et al.

108. Ota R, So K, Tsuda M, Higuchi Y, Yamashita F (2021) Prediction of HIV drug resistance
based on the 3D protein structure: proposal of molecular field mapping. PLoS One 16(8
August):1–15. https://doi.org/10.1371/journal.pone.0255693
109. Sheik Amamuddy O, Bishop NT, Tastan Bishop Ö (2017) Improving fold resistance prediction
of HIV-1 against protease and reverse transcriptase inhibitors using artificial neural networks.
BMC Bioinf 18(1):1–7. https://doi.org/10.1186/s12859-017-1782-x
Chapter 7
Recent Insight of the Emerging Severe
Fever with Thrombocytopenia Syndrome
Virus: Drug Discovery, Therapeutic
Options, and Limitations

Shilpa Chatterjee, Arindam Maity, and Debanjan Sen

Abstract Severe fever with thrombocytopenia syndrome virus (SFTSV) also known
as Dabie bandavirus of the family Phenuiviridae is a negative-strand RNA virus and
a tick-borne virus. Replication of SFTSV into systemic circulation and occurrence
of viremia cause cytokine storm and T-cell overstimulation. The event of viremia-
induced thrombocytopenia causes reduced platelet count and splenic macrophages,
followed by endothelial damages and compromised immune system that cause multi-
organ damages. Limited options for specific anti-SFTSV drugs pose significant chal-
lenges associated with clinical management of SFTSV infection. This book chapter
chiefly emphasizes upon the genetic diversity, geographical distribution, pathogen-
esis associated with various clinical aspects like symptoms, diagnosis, and available
clinical management options. In addition, current research linked with anti-SFTSV
drug development is comprehensively portrayed in this review.

Keywords Severe Fever with Thrombocytopenia Syndrome virus (SFTSV) ·

Dabie bandavirus · Huaiyangshan Banyangvirus · Phenuiviridae · Clinical
symptoms of SFTSV infection · Clinical diagnosis of SFTSV · SFTS L protein ·
SFTSV drug target · Molecular Docking · Molecular Dynamics

S. Chatterjee
Department of Biomedical Science, College of Medicine, Chosun University, Gwangju, Republic
of Korea
e-mail: [email protected]
A. Maity
Department of Pharmaceutical Technology, JIS University, Kolkata, India
D. Sen (B)
Department of Pharmaceutical Technology, BCDA College of Pharmacy & Technology, 78
Jessore Road, Hridaypur, Kolkata 700127, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 195
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_7
196 S. Chatterjee et al.

7.1 Introduction

Dabie bandavirus, also called severe fever with thrombocytopenia syndrome virus
(SFTSV), is a tick-borne virus of the genus Bandavirus, belonging to the family
Phenuiviridae, order Bunyavirales [1]. Synonymously SFTSV is also known as
Huaiyangshan Banyangvirus. The clinical condition caused by SFTSV is known
as severe fever with thrombocytopenia syndrome (SFTS) [1]. Although the new
nomenclature was accepted by the International Committee on Taxonomy of Viruses
(ICTV), SFTSV is the most widely used term in the scientific community. The
SFTSV can be classified as follows: Realm (Riboviria), Kingdom (Orthornavirae),
Phylum (Negarnaviricota), Class (Ellioviricetes), Order (Bunyavirales), Family
(Phenuiviridae), Genus (Bandavirus), Species (Dabie bandavirus).
Figure 7.1 describes the schematic diagram of SFTSV. SFTSV is a negative-strand
RNA that has been divided into large (L), medium (M), and small (S) segments [2].
The RNA-dependent RNA polymerase (RdRp), which acts as a viral transcriptase/
replicase, is encoded by the L segment. The M segment codes for a membrane
protein precursor that matures into the envelope’s two glycoproteins, Gn and Gc.
The S segment is a two-protein ambisense RNA; the antisense RNA encodes Np
and the sense RNA encodes NSs. Np is involved in the encapsidation of viral RNA
and the creation of the RNP complex, while NSs interfere with the synthesis of host
interferon [3].
Aside from the fact that SFTSV-related health problems are becoming more preva-
lent in people, the pathogenesis of the SFTS virus in humans is still unknown, and no
cure for the virus exists. Avoiding tick bites is a simple approach to protect ourselves
from infection. As a result, this disease has developed to cause major health problems
in humans in a variety of places around the world. In this chapter, we have discussed
about SFTS disease and its causative agent, epidemiology, pathogenesis, diagnosis,
and recent development in the treatment.

Fig. 7.1 Schematic diagram of SFTSV

7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 197

7.2 Geographical Distribution and Its Genetic Diversity

The first case of SFTS was recorded in Henan and Hubei provinces of China in 2009,
and it quickly spread to neighboring provinces in the country’s central, eastern, and
northeastern regions [2]. In 2012, SFTS cases were also recorded in Japan and Korea,
as well as Vietnam and Taiwan [4–7]. The mechanisms behind the spread of SFTS
are unknown; however, the spread of emerging viruses is commonly attributed to
two main mechanisms: increased contact between wildlife and human populations
and geographical spread of the hematophagous arthropod vector or their vertebrate
host outside of the endemic area. Ticks carrying the parasite H. longicornis are a
widespread parasite of migratory birds that breed and travel between endemic sites
in China, Korea, and Japan [2]. Furthermore, the Asia–Pacific range of H. longicornis
corresponds to the migration route of birds in the East Asian-Australasian flyway.
This suggests that migrating birds are involved in the spread of H. longicornis [8].

7.3 Mechanism and Pathogenesis of SFTSV

1—SFTSV human transmission via tick bite. 2—SFTSV carrying tick target nearest
lymph nodes generating impaired immune response via B-cell differentiation. 3—
further replication of SFTSV into systemic circulation and occurrence of viremia
causing cytokine storm and T-cell overstimulation. 4—viremia induced thrombo-
cytopenia; causing reduced platelet count and splenic macrophages. 5—endothelial
damages and compromised immune system cause multi-organ damages (Fig. 7.2).
The virus propagates inside the cytoplasm of host cells after infection by SFTSV,
and the release of RNPs by SFTSV begins transcription catalyzed by viral RdRp.
Complementary RNAs (cRNAs) and viral RNAs are produced because of viral RNA
replication aided by protein synthesis (vRNAs). The duplication of all three segments
(S, M, and L) varies. Due to the interaction between viral protein N and RdRp, the
produced cRNAs and vRNAs are then packed in RNP. RNPs that have just been
produced are employed to make viral mRNA and protein [9].
The etiology of SFTSV is unknown, however according to bunyavirus
pathogenicity, SFTSV suppresses the host’s immune response, resulting in intense
virus proliferation and organ failure. After investigating SFTSV patients, it was
shown that CD3-positive and CD4-positive T lymphocytes, which play a role in
immune function, are in lower numbers than normal, and the number of natural
killer cells is higher, especially in the acute and severe stages of the infection [10].
By stimulating the patient’s health situation, immune function suppression promotes
the spread of secondary infection. Natural killer cells conduct immunoregulatory
tasks by generating cytokines such as interferon, tumor necrosis factor (TNF), inter-
leukin 10, and granulocyte colony-stimulating factors (G-CSF). These cytokines are
proportional to the severity of the condition. Inflammatory cytokines also play a role
in the pathophysiology of viral infections, and some pro-inflammatory cytokines are
198 S. Chatterjee et al.

Fig. 7.2 Mechanism of action of SFTSV pathogenesis

overexpressed in the cytokine pool, indicating a severe form of SFTS [11]. Inter-
feron is produced by the innate immune system to guard against viruses, but people
with SFTS lack interferon. In SFTSV-infected monocytes, all interferon-related tran-
scription factors are moderately upregulated, whereas upstreaming molecules like
TNF-receptor-associated factors 3, 6 and antiviral signaling protein of mitochon-
dria are either unaffected or decreased, further inhibiting interferon induction [12].
Unbalanced cytokines such as interleukins 6 and 10, interleukin-1 receptor antago-
nist, G-CSF, interferon-inducible protein, and monocyte chemotactic protein 1 have
exhibited three distinct patterns in fatal instances of SFTS compared to nonfatal
cases. Platelet-derived growth factor (PDGF) and regulated on activation and gener-
ally expressed by T-cells (RANTES) expression, on the other hand, are low. All of
these cytokines recover to normal levels throughout the convalescent phase. Only
in fatal cases as well as in the convalescent phase of survivors does the expres-
sion of interleukins 1 and 8 and inflammatory proteins 1 and 1 of macrophages rise
[13]. Fever symptoms are linked to elevated TNF in SFTS patients, which works on
the endothelium, boosts vasodilating chemicals, nitric oxide synthase, and increases
vascular permeability [14]. SFTS virus attaches to platelets, which are recognized
and destroyed by circulating macrophages in the spleen, resulting in thrombocy-
topenia [15]. SFTSV can replicate mostly in reticular cells, although it can also
multiply in other cells [16]. Several organ failures arose from SFTSV’s targeting
of multiple organs. Therefore, starting with a feverish illness in the acute phase,
the patient develops multiple organ failure (severe form) and eventually dies (fatal
form). In fatal human SFTS cases, SFTSV infects B-cells, lymphocytes, and several
7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 199

lymphoid or nonlymphoid organs, including the blood, spleen, liver, adrenal glands,
gut, heart, lungs, and kidneys [17].

7.4 Clinical Symptoms

The incubation period for SFTS can extend anywhere from 5 to 14 days, depending on
viral levels and the point of infection [18]. The regular tick bite skin markings do not
have eschar, which is typical of scrub typhus patients [19]. Fever, gastrointestinal
symptoms (e.g., nausea, vomiting, stomach pain, and diarrhea), and neurological
symptoms (e.g., altered mental status) characterize the majority of patients [2, 18].
Thrombocytopenia (100,000/mm3 ) and leukopenia (4000/mm3 ) were found in the
majority of SFTS patients, along with raised alanine aminotransferase (ALT), aspar-
tate aminotransferase (AST), and alkaline phosphatase (ALP) levels and acute renal
injury. Lactate dehydrogenase (LDH) and ferritin levels rise as well, as does the
activated partial thromboplastin time (aPTT), as well as proteinuria with or without
hematuria [2, 18, 20]. Cardiomegaly with or without pericardial effusion and patchy
consolidation with ground-glass opacity (GGO) are the most common findings on
chest radiographs in patients with SFTS, which aids in the early differentiation
of SFTS from scrub typhus, which is characterized by interstitial pneumonia on
chest radiographs [21]. During the second week of illness, most patients with severe
SFTS die from MOF, which includes acute renal injury, myocarditis, arrhythmia, and
meningoencephalitis [22, 23]. The average time from the commencement of illness
to death is 9 days [24]. The fatality rate for SFTS varies between 6 and 21% [25–
27]. Advanced age, altered mental condition, higher serum LDH and AST levels,
prolonged aPTT, and high viral RNA loads in the serum are all poor prognostic
markers [26–30]. Similar to cytokine, LDH, AST, and blood urea nitrogen (BUN)
levels, viral RNA load has been found to provide useful information for treatment
strategies or the prognosis of patients with SFTS [31]. These findings are in line with
what has been observed in humans.

7.5 Diagnosis

SFTS is a disorder that is difficult to detect if medical personnel are unaware of it.
Fever, low platelet counts, and low white blood cell counts are common symptoms in
patients. If the patients have a history of tick bites in endemic locations such as central
and eastern China, rural South Korea, or southern Japan, SFTS should be considered.
The importance of early diagnosis of SFTSV infection for patient survival cannot
be overstated. Because the clinical indications of SFTS are nonspecific, laboratory
confirmation is required; also, other tick-borne diseases such as scrub typhus and
anaplasmosis generate comparable symptoms [19, 22]. For laboratory diagnosis of
SFTS, reverse transcriptase (RT) real-time PCR for the detection of viral RNA in
200 S. Chatterjee et al.

the serum during the first week of illness is a very sensitive and specific diagnostic
technique [32]. In the acute phase and for up to 20 days after the onset of symptoms,
viral RNA can be found in the blood; however, analyzing serum samples within
2 weeks of the onset of sickness is recommended [32]. RT-PCR approaches based on
the nucleotide sequence of SFTSV strains reported in China may be less susceptible to
diagnoses of the SFTS lineage identified in other countries due to significant genetic
differences among SFTSV inhabitants. To overcome the aforementioned obstacle,
Yoshikawa et al. devised a sensitive and specific conventional one-step RT-PCR
method as well as a quantitative one-step RT-PCR that can detect both strains [33].
In addition, a number of PCR approaches are being developed to diagnose SFTSV
more readily and swiftly. Huang et al. devised a reverse transcription-loop-mediated
isothermal amplification (RT-LAMP) approach that has 99% sensitivity and 100%
specificity for detecting novel bunyaviruses [34]. Baek et al. also demonstrated that
RT-LAMP may provide quick diagnosis in 30–60 min with a sensitivity 10 times
higher than traditional RT-PCR [35]. IFA or an enzyme-linked immunosorbent assay
(ELISA) are effective diagnostic methods for detecting viral-specific IgM and IgG
in the serum 7 days after the onset of the disease; SFTS is diagnosed when IgM
antibodies are detected, IgG antibody seroconversion is observed, or the antibody
titer increases by at least fourfold [18]. However, IFA sensitivities for IgM and
IgG detection after 2 weeks following onset of symptoms are 32–62% and 63–76%,
respectively, while ELISA sensitivities are 53–62% and 58–86% [36]. As a result, IFA
or ELISA may be insufficient for SFTS diagnosis in the early stages. Hemorrhagic
fever with renal syndrome (HFRS), severe dengue fever, thrombocytopenic purpura
(TTP), leptospirosis, human granulocytic anaplasmosis (HGA), and Lyme disease
are all viral infections with hemorrhagic fever. Patients with these disorders have
clinical symptoms and test results that are comparable to those who have SFTS.
As a result, in locations where illnesses coexist with SFTS, differential diagnosis is
critical (e.g., South Korea, China, and Japan). Scrub typhus and SFTS, in example,
cause identical clinical symptoms and test findings in endemic locations. When a
score of 2 was obtained after the evaluation of four variables (i.e., altered mental
status, leukopenia, prolonged aPTT, and normal C-reactive protein levels), all of
which weighed one point, Kim et al. proposed a scoring system that showed 100%
sensitivity and 97% specificity [19]. Li et al. proposed a multiplex real-time RT-PCR
assay to undertake successful screening for early SFTS diagnosis and to differentiate
it from other diseases (such as those caused by the Hantaan, Seoul, and dengue
viruses) in the acute phase to more easily and rapidly identify the infections [37].
Virus isolation for laboratory diagnosis, on the other hand, is currently challenging
to implement in the clinic because it requires a BSL3 laboratory and takes 2–5 days.
7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 201

7.6 SFTS Therapeutic Options

There have been no prospective randomized studies on treatment options that have
shown to be beneficial in the treatment of SFTS. Symptoms such as fever, diarrhea,
dehydration, bleeding propensity, and shock are treated with conservative measures
such as hydration, transfusion, and the use of antipyretics, inotropic drugs, and G-
CSF. However, in the acute stage, rapidly advancing cases of the disease are difficult
to treat appropriately; many patients with severe SFTS are thought to have developed
sepsis or septic shock due to verified MOF before being recognized. As a result, early
detection of SFTS is critical. Due to the difficulties of therapy and the high mortality
rate of SFTS, various treatments have been tried. In the following paragraphs, we
will discuss therapy approaches that have been presented.
Antiviral Drugs

1. Favipiravir

Toyama Chemical Co., Ltd. developed and produced favipiravir (T-705), which has
broad antiviral activity against RNA viruses including as influenza, arenaviruses,
bunyaviruses, West Nile virus, yellow fever virus, and foot-and-mouth disease
virus [38]. Host enzymes convert favipiravir to its active form, ribofuranosyl-5-
triphosphate, which inhibits viral RNA polymerase in the host cells. In vitro favipi-
ravir resistance has only been reported in a few cases [39, 40]. Furthermore, in Vero
cells [41], the IC90 of favipiravir (22 μM) was lower than that of ribavirin (263 μM)
[42]. Animal models have been used to test the efficacy of favipiravir in vivo. Favipi-
ravir, given intraperitoneally (i.p.) at doses of 60 or 300 mg/kg/day for 5 days, totally
protected mice from death caused by SFTSV infection, with only a minor weight
loss [41]. When favipiravir treatment began on or before 3 days after infection, all
mice survived, whereas animals treated at 4 and 5 days after infection had 83% and
50% survival, respectively [41]. These findings suggested that favipiravir could be
used as a preventative as well as a treatment for SFTSV infections. In most cases,
favipiravir is taken orally by humans. In a mouse model, favipiravir given orally
(p.o.) had equal efficacy to favipiravir given intravenously (i.p). [43]. Furthermore,
in a STAT2 deletion golden Syrian hamster model, therapy with favipiravir (300 or
150 mg/kg/day) offered complete protection against a deadly SFTSV challenge [44].
2 Ribavirin

Ribavirin is a nucleotide analog having broad-spectrum antiviral function against

various viruses. It can be given intravenously, orally, or by a nebulizer [45]. Ribavirin
has both direct and indirect modes of action against viruses, including the suppres-
sion of inosine monophosphate dehydrogenase and immunomodulatory effects [46].
It has also been investigated whether ribavirin can be used to treat SFTS sufferers. A
study on the effects of ribavirin on SFTSV was published in 2017 by Lee et al. which
reported that ribavirin decreased cytopathic effects and replication of SFTSV at an
202 S. Chatterjee et al.

IC50 ranging from 3.69 to 8.72 g/mL [47]. So far, several studies have performed to
identify the effects of ribavirin on SFTS but most of these are combination thera-
pies along with ribavirin used for SFTS treatment. Additionally, anemia and hyper-
amylasemia are two adverse effects of ribavirin that have been documented [48].
Therefore, ribavirin administration is not a proven viable therapy option [49, 50].
3 Hexachlorophene

Yuan et al. (2019) screened an FDA-approved drug library containing 1528 drug
compounds and found five that inhibited SFTSV replication at 10 μM concentra-
tions, including two antibacterial and antifungal disinfectants (hexachlorophene and
triclosan), a multi-kinase inhibitor for the treatment of advanced solid organ tumors
(regorafenib), and a small molecule agonist of the C-mannosylation of thrombo
(broxyquinoline) [51]. Hexachlorophene was the most potent of them all, with an
IC50 of 1.3 ± 0.3 μM (RNA load) and 2.6 ± 0.14 μM (plaque reduction) and the
highest selectivity index (50% cytotoxic concentration [CC50]/IC50, 18.7), which
was lower than the other four antiviral medicines. Furthermore, the findings revealed
that hexachlorophene treatment inhibited SFTSV entrance while having no effect on
virus-host cell adhesion or virus infectivity [51]. Hexachlorophene was anticipated
to attach to the deep hydrophobic pocket between domains I and III of the SFTSV
Gc glycoprotein, causing cell membrane fusion to be disrupted. Hexachlorophene
is an antibacterial chemical that is commonly found in soaps and scrubs, as well as
an experimental cholinesterase inhibitor [52]. In vitro, hexachlorophene suppressed
the viral replication of a coronavirus linked to severe acute respiratory syndrome by
blocking 3C-like protease, which is required for the virus’s lifecycle [52].

4 2' -Fluoro-2' -deoxycytidine

The nucleoside inhibitor 2' -fluoro-2' -deoxycytidine (2' -FdC) is employed in anti-
cancer medications. Borna virus [53], Lassa virus [54], Crimean-Congo hemorrhagic
fever virus [55], influenza virus [56], and herpesviruses are among the RNA and DNA
viruses that it suppresses in vitro [57] 2' -FdC has been reported to have antiviral action
against a variety of bunyaviruses, including La Crosse virus, Maporal virus, Punta
Toro virus, Rift Valley fever virus, San Angelo virus, Heartland virus, and SFTSV,
according to [58]. In an in vitro test, the IC90 of 2' -FdC against SFTSV was 3.7 μM.
A 100 mg/kg/day therapy with 2' -FdC was 100% protective against death caused by
SFTSV in an in vivo research utilizing IFNAR/mice. However, after SFTSV inoc-
ulation, all mice treated with 2' -FdC lost a significant amount of weight, whereas
mice treated with favipiravir lost very little weight, suggesting that favipiravir was
more efficient than 2' -FdC in limiting morbidity during infection [58].

5 Calcium Channel Blockers

Calcium channel blockers (CCBs) lower intracellular Ca2+ levels and are commonly
used to treat hypertension, angina, and supraventricular arrhythmias, among other
cardiovascular conditions. Antiviral activity of CCBs has recently been reported
against ebolavirus, marburgvirus, Junn virus, West Nile virus, and Japanese
7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 203

encephalitis virus [59–63]. The CCBs benidipine hydrochloride and nifedipine were
discovered as inhibitors of SFTSV replication in vitro by limiting viral internal-
ization and lowering genome replication during the post-entry phase, according to
a screening of 700 FDA-approved medicines [64]. The viral binding, fusion, and
budding were not affected by this mechanism. Treatment with benidipine hydrochlo-
ride or nifedipine decreased SFTSV replication by lowering virus-induced Ca2+
influx, according to the findings of an in vitro investigation. In C57BL/6 mice and
humanized mouse models, the anti-SFTSV effect of these two CCBs was further
investigated, demonstrating treatment results of reduced viral load, improved platelet
count, and lower fatality rate in the humanized mouse model. Notably, nifedipine
is one of the most commonly prescribed medications in China for the treatment of
hypertension and atherosclerosis. As a result, Li et al. (2019) conducted a retro-
spective clinical investigation on a large cohort of 2087 SFTS patients, including
83 nifedipine-treated patients who received nifedipine before admission and during
hospitalization, 48 non-nifedipine-treated patients who received nifedipine before
admission but not during hospitalization, and 249 general SFTS patients who did
not receive nifedipine at all [64]. The case fatality rate in the nifedipine-treated
group (3.6%) was less than half that of the overall SFTS group (19.7%) or the
non-nifedipine-treated group (20.8%) [64]. In contrast to ribavirin, nifedipine-
treated patients with a high viral load (> 106 copies/mL) had a significantly lower-
case fatality rate (2.4%) as compared to general SFTS patients (29%) and non-
nifedipine-treated patients (34.5%). Hematemesis was shown to be less common
in the nifedipine-treated group, which is one of the hemorrhagic symptoms that is
closely linked to death. The authors demonstrated the inhibitory effect of benidipine
hydrochloride or nifedipine in cultured cells in an animal model in this article. Most
importantly, it was discovered that nifedipine treatment boosted viral clearance and
clinical recovery.
6 Caffeic Acid

Caffeic acid (CA) is a polyphenol chemical component connected to coffee that

can be found in a variety of plants, including coffee beans. Chlorogenic acid, the
ester of caffeic acid, is found in 70–350 mg per cup of coffee [65]. It has a number
of biological effects, including cancer cell suppression and antiviral activities [66–
71]. In an in vitro test employing Huh 7.5.1–8 cells, a highly tolerant derivation of
human hepatoma Huh7 cells, found that CA suppressed SFTSV replication dose-
dependently [72]. CA had an IC50 of 48 μM and a CC50 of 7.6 mM against SFTSV.
Surprisingly, pretreatment of SFTSV with CA before inoculation lowered the virus
copy number in the supernatant of infected cells at 72 h after infection, and the
inhibitory impact was greatly diminished when the cells were treated with CA after
SFTSV inoculation. As a result, the scientists hypothesized that CA worked mostly
204 S. Chatterjee et al.

on viral particles or influenced the early stages of SFTSV infection, while it might
also limit viral genome replication in host cells.
7 Amodiaquine
Amodiaquine, a new antimalarial medication, has been shown to have antiviral effects
against ebolavirus, dengue virus, and zika virus [72–76]. The mechanism of amodi-
aquine’s inhibitory effect against malaria and those viruses is unknown. Amodi-
aquine and other halogen compounds (fluorine, bromine, and iodine) were tested
against SFTSV replication in vitro by [77]. The IC50 for fluorine, bromine, and
iodine, respectively, was 36.6, 31.1, and 15.6 μM for fluorine, bromine, and iodine
compound. Amodiaquine was found to be a selective inhibitor of SFTSV replica-
tion among the drugs examined. Amodiaquine had a CC50 of >100 and an IC50 of
19.1 μM, respectively. Amodiaquine IC50 was lower than ribavirin’s (40.1 μM) and
favipiravir’s (25.0 μM).
8 IFN-γ
Type II IFNs only have one member, IFN-γ . By modulating antigen processing and
presentation pathways, it encourages macrophages and dendritic cells to provide
direct antimicrobial activity. Activated T-cells and activated natural killer cells
were assumed the only important sources of IFN-γ , but under certain conditions,
macrophages and dendritic cells can also be driven to create IFN-γ in vitro [78].
IFN-γ plays a crucial function in viral infection because it can directly increase
the development of several putative antiviral IFN-stimulating proteins via STAT1
signaling.
9. Monoclonal antibodies (Mab)
Monoclonal antibodies are regarded as new therapeutic agents for SFTS across a
variety of treatment options. According to Guo et al., monoclonal antibodies worked
to neutralize SFTSV infection in Vero cells by attaching to a linear epitope in
the glycoprotein Gn’s ectodomain [79]. This neutralizing activity results from the
suppression of interactions between glycoprotein Gn and cellular receptors, which
prevents viral cell attachment. Additionally, Kim et al. revealed that their chosen
antibody was reactive to the SFTSV’s envelope glycoprotein Gn and protected 80%
of mice and host cells [80], indicating that monoclonal antibodies could be able to
defend against SFTSV. These findings imply that monoclonal antibodies may provide
SFTS patients a promising therapy alternative.

7.7 Structure-Based Drug Design Approach Guided

Identification of Potential Binders

The viral L protein can be considered as one of the emerging targets for devel-
oping therapeutic agents. The L protein synthesizes three different RNA species
during the viral replication [81] and pose significant importance. The cap binding
7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 205

Fig. 7.3 Potential SFTS-L protein binder identified by computational methods

domain of the SFTS-L protein extensively utilized to identify potential binders. The
drug re-purposing approach identified Zaltoprofen as a potential SFTS-L protein
binder Fig. 7.3 [82]. The Phe1703, Tyr1719; Gln1707, Asp1771, Leu1772; Pro1706,
Ile1738, and Ile1774; Phe1703 and Tyr1719; Phe1703, Gln1707, Asp1771, and
Trp1725; Pro1706, Ile1738, Ile1774, Leu1768, Leu1772 residues of SFTS-L protein
(PDB ID:6XYA) are the crucial amino acids present in the binding site. The molec-
ular docking followed by molecular dynamics investigation identified β-sesqui-
phellan-drene as a binder of membrane glycoprotein polyprotein of SFTS virus
[83].
Few more chemical compounds like Bromfenac, Cinchophen, and Elliptinium
depict stable binding with SFTS-L protein with acceptable docking score. The
molecular dynamics study of these compounds bound systems was conducted.
The researchers calculated various parameters like RMSD, RMSF, Radius of
Gyration (Rg), Solvent accessible surface area, etc., from the molecular dynamic’s
trajectory can be found in Fig. 7.4. Each holo-protein (ligand bound protein) depicts
lower RMSD, RMSF, and Rg profile in compared to apo-protein (ligand free protein).
The RMSD values were found to be less then 3.0 Å. ´ RMSD profile for a glob-
´ indicated a stable system [84]. On the basis of RMSD
ular protein less then 3.0 Å
profile associated with other parameters, it can be stated these chemical compounds
identified from Drugbank Database exhibited stable binding with SFTS-L protein.
Virtual screening of Indian natural products presents in Indian medicinal plants
identified few potential hits against SFTS-L protein. The name of the hits are
Gamma-glutamylaspartic acid, 2' -Deoxymugineic acid, Traumatic acid, Betalamic
acid, Epoxyoleic acid, respectively [85]. Studies depict presence of divalent Mn2+
ions in the SFTS-L protein binding site required for the activity [85].
206 S. Chatterjee et al.

Fig. 7.4 Various parameters calculated from molecular dynamics trajectory. The figure is repro-
duced from Ref. [82] open access under a CC BY 4.0 license, https://creativecommons.org/licenses/
by/4.0/

7.8 Conclusion

The SFTSV can be considered as one of the most life-threatening infections. Till date
lack of appropriate medications pose significant emergence to develop new thera-
peutic agents. Unavailability of large number of chemical entities exhibiting distinct
anti-SFTS properties, structure-based drug design approach was largely adopted
over ligand-based drug design approach, to identify potential hits. Recent molecular
docking followed by molecular dynamics guided approach identified sets of hits
against SFTS-l protein. However, there is a huge scope to identify more relevant hits
against this virus in order to complete the journey of a small molecules from bench
to bed side.

References

1. ICTV. ICTV Taxonomy History: SFTS virus. https://talk.ictvonline.org/taxonomy/p/taxono

myhistory?taxnode_id=20141803&src=NCBI&ictv_id=20141803 (ICTV, 2020)
2. Yu X-J et al (2011) Fever with thrombocytopenia associated with a novel bunyavirus in China.
N Engl J Med 364:1523–1532
7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 207

3. Wiwanitkit S, Wiwanitkit V (2015) Acute viral hemorrhage disease: a summary on new viruses.
J Acute Dis 4:277–279
4. Takahashi T et al (2014) The first identification and retrospective study of severe fever with
thrombocytopenia syndrome in Japan. J Infect Dis 209:816–827
5. Kim K-H et al (2013) Severe fever with thrombocytopenia syndrome, South Korea, 2012.
Emerg Infect Dis 19:1892
6. Tran XC et al (2019) Endemic severe fever with thrombocytopenia syndrome. Vietnam Emerg
Infect Dis 25:1029
7. Lin T-L et al (2020) The first discovery of severe fever with thrombocytopenia syndrome virus
in Taiwan. Emerg Microbes Infect 9:148–151
8. Yun Y et al (2015) Phylogenetic analysis of severe fever with thrombocytopenia syndrome
virus in South Korea and migratory bird routes between China, South Korea, and Japan. Am J
Tropical Med Hyg. 93:468–474
9. Lei X-P, Liu M, Yu X (2015) Severe fever with thrombocytopenia syndrome and its pathogen
SFTSV. Microbes Infect 17:149–154
10. Sun L, Hu Y, Niyonsaba A et al (2013) Detection and evaluation of immunofunction of patients
with severe fever with thrombocytopenia syndrome. Clin Exp Med. https://doi.org/10.1007/
s10238-013-0259-0
11. Deng B, Zhang S, Geng Y et al (2012) Cytokine and chemokine levels in patients with severe
fever with thrombocytopenia syndrome virus. PLoS ONE 7:41365
12. Qu B, Qi X, Wu X et al (2012) Suppression of the interferon and NF-κB responses by severe
fever with thrombocytopenia syndrome virus. J Virol 86:8388–8401
13. Sun Y, Jin C, Zhan F et al (2012) Host cytokine storm is associated with disease severity of
severe fever with thrombocytopenia syndrome. J Infect Dis 206:1085–1094
14. Seynhaeve ALB, Vermeulen CE, Eggermont AMM, Hagen TLMT (2006) Cytokines and
vascular permeability: an in vitro study on human endothelial cells in relation to tumor necrosis
factor-alpha-primed peripheral blood mononuclear cells. Cell Biochem Biophys 44:157–169
15. Jin C, Liang M, Ning J et al (2012) Pathogenesis of emerging severe fever with thrombocy-
topenia syndrome virus in C57/BL6 mouse model. Proc Natl Acad Sci USA 109:10053–10058
16. Liu Q, Biao H, Si-Yang H, Feng W, Xing-Quan Z et al (2014) Severe fever with thrombocy-
topenia syndrome, an emerging tick-borne zoonosis. Lancet Infect Dis 14:763–772
17. Suzuki T, Sato Y, Sano K, Arashiro T, Katano H, Nakajima N, Morikawa S et al (2020) Severe
fever with thrombocytopenia syndrome virus targets B cells in lethal human infections. J Clin
Investig 130(2):799–812
18. Liu Q, He B, Huang SY, Wei F, Zhu XQ (2014) Severe fever with thrombocytopenia syndrome,
an emerging tick-borne zoonosis. Lancet Infect Dis 14:763–772
19. Kim MC, Chong YP, Lee SO, Choi SH, Kim YS, Woo JH, Kim SH (2018) Differentiation of
severe fever with thrombocytopenia syndrome from scrub typhus. Clin Infect
20. Kim UJ, Oh TH, Kim B, Kim SE, Kang SJ, Park KH, Jung SI, Jang HC (2017) Hyper-
ferritinemia as a diagnostic marker for severe fever with thrombocytopenia syndrome. Dis
Markers 2017:6727184
21. Yun JH, Hwang HJ, Jung J, Kim MJ, Chong YP, Lee SO, Choi SH, Kim YS, Woo JH, Kim
MY et al (2019) Comparison of chest radiographic findings between severe fever with throm-
bocytopenia syndrome and scrub typhus: single center observational cross-sectional study in
South Korea. Medicine 98:e17701
22. Miyamoto S, Ito T, Terada S, Eguchi T, Furubeppu H, Kawamura H, Yasuda T, Kakihana Y
(2019) Fulminant myocarditis associated with severe fever with thrombocytopenia syndrome:
a case report. BMC Infect Dis 19:266
23. Park SY, Kwon JS, Kim JY, Kim SM, Jang YR, Kim MC, Cho OH, Kim T, Chong YP, Lee
SO et al (2018) Severe fever with thrombocytopenia syndrome-associated encephalopathy/
encephalitis. Clin Microbiol Infect 24:432.e1–432.e4
24. Ding F, Zhang W, Wang L, Hu W, Soares Magalhaes RJ, Sun H, Zhou H, Sha S, Li S, Liu Q
et al (2013) Epidemiologic features of severe fever with thrombocytopenia syndrome in China,
2011–2012. Clin Infect Dis 56:1682–1683
208 S. Chatterjee et al.

25. Sun J, Lu L, Wu H, Yang J, Ren J, Liu Q (2017) The changing epidemiological characteristics
of severe fever with thrombocytopenia syndrome in China, 2011–2016. Sci Rep 7:9236
26. Choi SJ, Park SW, Bae IG, Kim SH, Ryu SY, Kim HA, Jang HC, Hur J, Jun JB, Jung Y et al
(2016) Severe fever with thrombocytopenia syndrome in South Korea, 2013–2015. PLoS Negl
Trop Dis 10:e0005264
27. Kato H, Yamagishi T, Shimada T, Matsui T, Shimojima M, Saijo M, Oishi K (2016) Epidemi-
ological and clinical features of severe fever with thrombocytopenia syndrome in Japan,
2013–2014. PLoS ONE 11:e0165207
28. Li H, Lu QB, Xing B, Zhang SF, Liu K, Du J, Li XK, Cui N, Yang ZD, Wang LY et al (2018)
Epidemiological and clinical features of laboratory-diagnosed severe fever with thrombocy-
topenia syndrome in China, 2011–2017: a prospective observational study. Lancet Infect Dis
18:1127–1137
29. Wang L, Wan G, Shen Y, Zhao Z, Lin L, Zhang W, Song R, Tian D, Wen J, Zhao Y et al
(2019) A nomogram to predict mortality in patients with severe fever with thrombocytopenia
syndrome at the early stage-A multicenter study in China. PLoS Negl Trop Dis 13:e0007829
30. Zhang YZ, He YW, Dai YA, Xiong Y, Zheng H, Zhou DJ, Li J, Sun Q, Luo XL, Cheng YL et al
(2012) Hemorrhagic fever caused by a novel Bunyavirus in China: pathogenesis and correlates
of fatal outcome. Clin Infect Dis 54:527–533
31. Hwang J, Kang JG, Oh SS, Chae JB, Cho YK, Cho YS, Lee H, Chae JS (2017) Molecular
detection of severe fever with thrombocytopenia syndrome virus (SFTSV) in feral cats from
Seoul Korea. Ticks Tick Borne Dis 8:9–12
32. Sun Y, Liang M, Qu J, Jin C, Zhang Q, Li J, Jiang X, Wang Q, Lu J, Gu W et al (2012) Early
diagnosis of novel SFTS bunyavirus infection by quantitative real-time RT-PCR assay. J Clin
Virol 53:48–53
33. Yoshikawa T, Fukushi S, Tani H, Fukuma A, Taniguchi S, Toda S, Shimazu Y, Yano K, Morim-
itsu T, Ando K et al (2014) Sensitive and specific PCR systems for detection of both Chinese
and Japanese severe fever with thrombocytopenia syndrome virus strains and prediction of
patient survival based on viral load. J Clin Microbiol 52:3325–3333
34. Huang XY, Hu XN, Ma H, Du YH, Ma HX, Kang K, You AG, Wang HF, Zhang L, Chen
HM et al (2014) Detection of new bunyavirus RNA by reverse transcription-loop-mediated
isothermal amplification. J Clin Microbiol 52:531–535
35. Baek YH, Cheon HS, Park SJ, Lloren KKS, Ahn SJ, Jeong JH, Choi WS, Yu MA, Kwon HI,
Kwon JJ et al (2018) Simple, rapid and sensitive portable molecular diagnosis of SFTS virus
using reverse transcriptional loop-mediated isothermal amplification (RT-LAMP). J Microbiol
Biotechnol 28:1928–1936
36. Ra SH, Kim MJ, Kim MC, Park SY, Park SY, Chong YP, Lee SO, Choi SH, Kim YS, Lee KH
et al (2020) Kinetics of serological response in patients with severe fever with thrombocytopenia
syndrome. Viruses 13:6
37. Li Z, Qi X, Zhou M, Bao C, Hu J, Wu B, Wang S, Tan Z, Fu J, Shan J et al (2013) A two-tube
multiplex real-time RT-PCR assay for the detection of four hemorrhagic fever viruses: Severe
fever with thrombocytopenia syndrome virus, Hantaan virus, Seoul virus, and dengue virus.
Arch Virol 158:1857–1863
38. Furuta Y, Takahashi K, Shiraki K, Sakamoto K, Smee DF, Barnard DL et al (2009) T-705
(favipiravir) and related compounds: novel broad-spectrum inhibitors of RNA viral infections.
Antiviral Res 82:95–102. https://doi.org/10.1016/j.antiviral.2009.02.198
39. Delang L, Guerrero NS, Tas A, Quérat G, Pastorino B, Froeyen M et al (2014) Mutations
in the chikungunya virus non-structural proteins cause resistance to favipiravir (T-705), a
broad-spectrum antiviral. J Antimicrob Chemother 69:2770–2784. https://doi.org/10.1093/jac/
dku209
40. Goldhill DH, Te Velthuis AJW, Fletcher RA, Langat P, Zambon M, Lackenby A et al (2018) The
mechanism of resistance to favipiravir in influenza. Proc Natl Acad Sci USA 115:11613–11618.
https://doi.org/10.1073/pnas.1811345115
41. Tani H, Fukuma A, Fukushi S, Taniguchi S, Yoshikawa T, Iwata-yoshikawa N et al (2016)
Efficacy of T-705 (Favipiravir) in the treatment of infections with lethal severe fever with
7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 209

thrombocytopenia syndrome virus. mSphere 1:e00061–e00015. https://doi.org/10.1128/mSp

here.00061-15
42. Shimojima M, Fukushi S, Tani H, Yoshikawa T, Fukuma A, Taniguchi S et al (2014) Effects
of ribavirin on severe fever with thrombocytopenia syndrome virus in vitro. Jpn J Infect Dis
67:423–427. https://doi.org/10.7883/yoken.67.423
43. Tani H, Komeno T, Fukuma A, Fukushi S, Taniguchi S, Shimojima M et al (2018) Therapeutic
effects of favipiravir against severe fever with thrombocytopenia syndrome virus infection in a
lethal mouse model: dose-efficacy studies upon oral administration. PLoS ONE 13:e0206416.
https://doi.org/10.1371/journal.pone.0206416
44. Gowen BB, Westover JB, Miao J, Van Wettere AJ, Rigas JD, Hickerson BT et al (2017)
Modeling severe fever with thrombocytopenia syndrome virus infection in golden syrian
hamsters: importance of STAT2 in preventing disease and effective treatment with favipiravir.
J Virol 91:e01942-e11916. https://doi.org/10.1128/JVI.01942-16
45. Snell NJ (2001) Ribavirin—current status of a broad-spectrum antiviral agent. Expert Opin
Pharmacother 2:1317–1324
46. Graci JD, Cameron CE (2006) Mechanisms of action of ribavirin against distinct viruses. Rev
Med Virol 16:37–48. https://doi.org/10.1002/rmv.483
47. Lee MJ, Kim KH, Yi J, Choi SJ, Choe PG, Park WB, Kim NJ, Oh MD (2017) In vitro antiviral
activity of ribavirin against severe fever with thrombocytopenia syndrome virus. Korean J
Intern Med 32:731–737
48. Lu QB, Zhang SY, Cui N, Hu JG, Fan YD, Guo CT, Qin SL, Yang ZD, Wang LY, Wang HY
et al (2015) Common adverse events associated with ribavirin therapy for severe fever with
thrombocytopenia syndrome. Antivir Res 119:19–22
49. Oh WS, Heo ST, Kim SH, Choi WJ, Han MG, Kim JY (2014) Plasma exchange and ribavirin
for rapidly progressive severe fever with thrombocytopenia syndrome. Int J Infect Dis 18:84–86
50. Park I, Kim HI, Kwon KT (2017) Two treatment cases of severe fever and thrombocytopenia
syndrome with oral ribavirin and plasma exchange. Infect Chemother 49:72–77
51. Yuan S, Chan JFW, Ye ZW, Wen L, Tsang TGW, Cao J et al (2019) Screening of an FDA-
approved drug library with a two-tier system identifies an entry inhibitor of severe fever with
thrombocytopenia syndrome virus. Viruses 11:E385
52. Hsu JTA, Kuo CJ, Hsieh HP, Wang YC, Huang KK, Lin CPC et al (2004) Evaluation of metal-
conjugated compounds as inhibitors of 3CL protease of SARS-CoV. FEBS Lett 574:116–120
53. Bajramovic JJ, Volmer R, Syan S, Pochet S, Gonzalez-Dunia D (2004) 2' -fluoro-2' -
deoxycytidine inhibits Borna disease virus replication and spread. Antimicrob Agents
Chemother 48:1422–1425
54. Welch SR, Guerrero LW, Chakrabarti AK, McMullan LK, Flint M, Bluemling GR et al (2016)
Lassa and Ebola virus inhibitors identified using minigenome and recombinant virus reporter
systems. Antiviral Res 136:9–18
55. Welch SR, Scholte FEM, Flint M, Chatterjee P, Nichol ST, Bergeron É et al (2017) Identification
of 2' -deoxy-2' -fluorocytidine as a potent inhibitor of Crimean-Congo hemorrhagic fever virus
replication using a recombinant fluorescent reporter virus. Antiviral Res 147:91–99
56. Kumaki Y, Day CW, Smee DF, Morrey JD, Barnard DL (2011) In vitro and in vivo efficacy of
fluorodeoxycytidine analogs against highly pathogenic avian influenza H5N1, seasonal, and
pandemic H1N1 virus infections. Antiviral Res 92:329–340
57. Wohlrab F, Jamieson AT, Hay J, Mengel R, Guschlbauer W (1985) The effect of 2' -fluoro-2' -
deoxycytidine on herpes virus growth. Biochim Biophys Acta 824:233–242
58. Smee DF, Jung KH, Westover J, Gowen BB (2018) 2' -Fluoro-2' -deoxycytidine is a broad-
spectrum inhibitor of bunyaviruses in vitro and in phleboviral disease mouse models. Antiviral
Res 160:48–54. https://doi.org/10.1016/j.antiviral.2018.10.013
59. Sakurai Y, Kolokoltsov AA, Chen CC, Tidwell MW, Bauta WE, Klugbauer N et al (2015) Two-
pore channels control Ebola virus host cell entry and are drug targets for disease treatment.
Science 347:995–998. https://doi.org/10.1126/science.1258758
60. Dewald LE, Dyall J, Sword JM, Torzewski L, Zhou H, Postnikova E et al (2018) The calcium
channel blocker bepridil demonstrates efficacy in the murine model of marburg virus disease.
J Infect Dis 22:S588–S591
210 S. Chatterjee et al.

61. Lavanya M, Cuevas CD, Thomas M, Cherry S, Ross SR (2013) siRNA screen for genes that
affect Junín virus entry uncovers voltage-gated calcium channels as a therapeutic target. Sci
Transl Med 5:204ra131
62. Scherbik SV, Brinton MA (2010) Virus-induced Ca2+ influx extends survival of west nile
virus-infected cells. J Virol 84:8721–8731. https://doi.org/10.1128/JVI.00144-10
63. Wang S, Liu Y, Guo J, Wang P, Zhang L, Xiao G et al (2017) Screening of FDA-approved
drugs for inhibitors of Japanese encephalitis virus infection. J Virol 91:e01055-e11017
64. Li H, Zhang LK, Li SF, Zhang SF, Wan WW, Zhang YL et al (2019) Calcium channel blockers
reduce severe fever with thrombocytopenia syndrome virus (SFTSV) related fatality. Cell Res
29:739–753
65. Clifford MN (1999) Chlorogenic acids and other cinnamates–nature, occurrence and dietary
burden. J Sci Food Agric 79:362–372
66. Tang H, Yao X, Yao C, Zhao X, Zuo H, Li Z (2017) Anti-colon cancer effect of caffeic acid
p-nitro-phenethyl ester in vitro and in vivo and detection of its metabolites. Sci Rep 7:7599
67. Bułdak RJ, Hejmo T, Osowski M, Bułdak Ł, Kukla M, Polaniak R et al (2018) The impact of
coffee and its selected bioactive compounds on the development and progression of colorectal
cancer in vivo and in vitro. Molecules 23:E3309
68. Wang GF, Shi LP, Ren YD, Liu QF, Liu HF, Zhang RJ et al (2009) Anti-hepatitis B virus
activity of chlorogenic acid, quinic acid and caffeic acid in vivo and in vitro. Antiviral Res
83:186–190
69. Utsunomiya H, Ichinose M, Ikeda K, Uozaki M, Morishita J, Kuwahara T et al (2014) Inhibition
by caffeic acid of the influenza a virus multiplication in vitro. Int J Mol Med 34:1020–1024
70. Ding Y, Cao Z, Cao L, Ding G, Wang Z, Xiao W (2017) Antiviral activity of chlorogenic acid
against influenza A (H1N1/H3N2) virus and its inhibition of neuraminidase. Sci Rep 7:1–11
71. Langland J, Jacobs B, Wagner CE, Ruiz G, Cahill TM (2018) Antiviral activity of metal chelates
of caffeic acid and similar compounds towards herpes simplex, VSV-Ebola pseudotyped and
vaccinia viruses. Antiviral Res 160:143–150. https://doi.org/10.1016/j.antiviral.2018.10.021
72. Ogawa M, Shirasago Y, Ando S, Shimojima M, Saijo M, Fukasawa M (2018) Caffeic acid, a
coffee-related organic acid, inhibits infection by severe fever with thrombocytopenia syndrome
virus in vitro. J Infect Chemother 24:597–601
73. Gignoux E, Azman AS, De Smet M, Azuma P, Massaquoi M, Job D et al (2016) Effect of
artesunate.amodiaquine on mortality related to Ebola virus disease. N Engl J Med 374:23–32
74. Sakurai Y, Sakakibara N, Toyama M, Baba M, Davey RA (2018) Novel amodiaquine derivatives
potently inhibit Ebola virus infection. Antiviral Res 160:175–182
75. Boonyasuppayakorn S, Reichert ED, Manzano M, Nagarajan K, Padmanabhan R (2014)
Amodiaquine, an antimalarial drug, inhibits dengue virus type 2 replication and infectivity.
Antiviral Res 106:125–134
76. Balasubramanian A, Teramoto T, Kulkarni AA, Bhattacharjee AK, Padmanabhan R (2017)
Antiviral activities of selected antimalarials against dengue virus type 2 and Zika virus. Antiviral
Res 137:141–150
77. Baba M, Toyama M, Sakakibara N, Okamoto M, Arima N, Saijo M (2017) Establishment of
an antiviral assay system and identification of severe fever with thrombocytopenia syndrome
virus inhibitors. Antivir Chem Chemother 25:83–89
78. Thäle C, Kiderlen AF (2005) Sources of interferon-gamma (IFN-γ) in early immune response to
Listeria monocytogenes. Immunobiology 210:673–683. https://doi.org/10.1016/j.imbio.2005.
07.003
79. Guo X, Zhang L, Zhang W, Chi Y, Zeng X, Li X, Qi X, Jin Q, Zhang X, Huang M et al (2013)
Human antibody neutralizes severe fever with thrombocytopenia syndrome virus, an emerging
hemorrhagic Fever virus. Clin Vaccine Immunol 20:1426–1432
80. Kim KH, Kim J, Ko M, Chun JY, Kim H, Kim S, Min JY, Park WB, Oh MD, Chung J (2019)
An anti-Gn glycoprotein antibody from a convalescent patient potently inhibits the infection
of severe fever with thrombocytopenia syndrome virus. PLoS Pathog 15:e1007375
7 Recent Insight of the Emerging Severe Fever with Thrombocytopenia … 211

81. Vogel D, Thorkelsson SR, Quemin ERJ, Meier K, Kouba T, Gogrefe N, Busch C, Reindl S,
Günther S, Cusack S, Grünewald K, Rosenthal M (2020) Structural and functional characteri-
zation of the severe fever with thrombocytopenia syndrome virus L protein. Nucleic Acids Res
48:5749–5765. https://doi.org/10.1093/NAR/GKAA253
82. Chatterjee S, Kim CM, Kim DM (2021) Potential efficacy of existing drug molecules against
severe fever with thrombocytopenia syndrome virus: an in silico study. Sci Rep 11(1):1–8.
https://doi.org/10.1038/s41598-021-00294-7
83. Joshi A, Sunil Krishnan G, Kaushik V (2020) Molecular docking and simulation investigation:
effect of beta-sesquiphellandrene with ionic integration on SARS-CoV2 and SFTS viruses. J
Genetic Eng Biotech 18. https://doi.org/10.1186/S43141-020-00095-X
84. Chatterjee S, Maity A, Chowdhury S, Islam A, Muttinini RK, Sen D (nd) In silico analysis and
identification of promising hits against 2019 novel coronavirus 3C-like main protease enzyme.
https://doi.org/10.1080/07391102.2020.1787228
85. Vivek-Ananth RP, Sahoo AK, Srivastava A, Samal A (2022) Virtual screening of phytochemi-
cals from Indian medicinal plants against the endonuclease domain of SFTS virus L polymerase.
RSC Adv 12:6234–6247. https://doi.org/10.1039/D1RA06702H
Chapter 8
Computational Toxicological Aspects
in Drug Design and Discovery, Screening
Adverse Effects

Emilio Benfenati, Gianluca Selvestrel, Anna Lombardo, and Davide Luciani

Abstract Toxicological aspects represent a fundamental step in the process of drug

design and discovery. There are multiple platforms available, and recently freely
available tools provided results comparable with those obtained from the commercial
ones. We will present examples of models for the different endpoints which can be
used. In addition, the future perspectives are to take into account in an earlier stage
the adverse effects, in order to simplify the long process of drug design and discovery,
and to optimize the selection of preferable features present in a new pharmaceutical.
In this new vision, a more holistic approach can apply multiple methodologies and
not only the screening of the adverse effects.

Keywords Drug · In silico · Read-across · Toxicology · VEGAHUB

8.1 Introduction

The use of in silico models is a fundamental component of all areas of science for
decades. The use of computers offers unique opportunities and opens new avenues
in research, in the development of new substances and products, and it may help
our society in many ways. Here, we address the sector of the applications related to
the evaluation of toxicity and how the computation tools may help. We will address
these specific topics. (i) Which models can be used to address toxicological endpoints.
Here, we will speak about in silico models, as opposed to in vivo and in vitro models.
In silico model, however, is a broad term which is also used in other areas related to
pharmaceuticals, such as clinical studies. In our case, we focus on models for toxicity.
(ii) We will address read-across, particularly the approaches where computer models
are relevant—read-across can be also done manually, and indeed historically, this
was the way to proceed. (iii) We will discuss how these two non-testing methods, in

E. Benfenati (B) · G. Selvestrel · A. Lombardo · D. Luciani

Istituto di Ricerche Farmacologiche Mario Negri IRCCS, Milano, Italy
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 213
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_8
214 E. Benfenati et al.

silico models and read-across, should be combined, and in general how to address
a weight-of-evidence approach, for the purpose to get a robust evaluation of the
toxicity value for a substance, taking advantage of multiple, heterogeneous data.
(iv) We will extend our discussion to the case when the evaluation of the toxicity
goes beyond one single endpoint and tries to get a comprehensive view. (v) As a
further exploration of the potential that a substance may be risky, we will see how
to integrate in silico tools for hazard and exposure. (vi) The tools and approaches
related to safety by design will be presented with some examples. (vii) We will see
that beyond pre-built models, there are software packages offering the possibility to
develop specific models on purpose. (viii) Finally, we will derive conclusions and
mention the new perspectives which are expected in future. Within this discussion,
we will provide reference to existing in silico tools or future perspectives and will
often refer to the architecture of software packages as in VEGAHUB (www.vegahu
b.eu), to show examples.

8.2 Tools for Individual Endpoints

The computational models to cope with toxicological aspects are more and more
numerous and sophisticated. There are multiple tools which can predict a large set
of endpoints, as listed here:
. Mutagenicity (Ames test)
. In vitro micronucleus
. In vivo micronucleus
. Chromosomal aberration
. Carcinogenicity
. Developmental toxicity
. Reproductive toxicity
. Repeated-dose toxicity
. Acute systemic toxicity
. Skin sensitization
. Skin irritation
. Eye irritation
. Liver toxicity
. Cardiotoxicity
. Neurotoxicity
. Nephrotoxicity
. Endocrine disruption.
This list is not exhaustive and limited to human health effects; some points
include several endpoints. We will introduce an example where in silico models
are well established, and then we will touch on some other cases to mention only the
key aspects. The reader interested in a more systematic discussion on the specific
endpoints may refer to a recent book [1]. Let us consider the case of the models
8 Computational Toxicological Aspects in Drug Design and Discovery … 215

for mutagenicity, determined with the Ames test. There are tens of in silico models
commercially available and other tens of models free. In the case of their use for
pharmaceutical impurities, the International Conference on Harmonization (ICH)
M7 guideline [2] includes recommendations on the use of in silico models, and in
particular, methods based on two approaches should be used together: One approach
refers to the rules defined by experts codifying structural alerts (SAs) associated to
mutagenic effect, and the second approach should be based on statistical methods [3].
In this way, the authorities recommend using two orthogonal methods. The reason is
that none of them is considered perfect, while the two methods may cover different
aspects, and thus, it is safer to have multiple tools to identify possible reasons of
concern. Although we agree with this perspective, we notice further aspects to be
discussed. One point refers to the regulatory context. In another area, of industrial
substances, within the REACH regulation (the European regulation on Registration,
Evaluation, Authorization and Restriction of Chemicals) [4], this strategy is also
welcome, but it has been also mentioned that, ideally, it should be preferable to have
ten separate values, one for each of the ten conditions to be used with different strains
with and without metabolic activation. We will not address this point here.
The use of the models based on SAs is quite convincing because it “explains” the
reason for the adverse effect. The user should be aware that the simple presence of a
SA in the molecule is not sufficient to label a substance. Each SA is present in a certain
number of mutagenic substances, but there are also non-mutagenic substances which
contain the SA. In some cases, most of the substances are not mutagenic. Indeed,
depending on the SA, there may be false positives, i.e. substances which are predicted
mutagenic while they are not. Another critical aspect is that some SAs are based on
very few substances. Furthermore, the different in silico models based on the SAs
contain different numbers of SAs, and they overlap only partially; thus, there is no
agreement in the community of experts. Finally, there is not a complete list of the
SAs, and thus, a substance may be mutagenic even if it does not contain a SA (so far
identified at least).
For all these reasons, the authorities rather recommend an additional, empir-
ical approach, so that in silico model may identify hazardous, substances which may
escape the identification of the adverse effect using SAs. We will come back on the
ways to integrate multiple values from different models later.
Of course, models are different according to the method employed. The models
may also differ with respect to other perspectives, and indeed we mentioned tens of
models for mutagenicity. In the case of the expert-based models, the differences are
quite subjective, since the rules have been codified by human experts, based on their
personal opinion, the paper they reviewed, the assumptions done. For instance, the
model derived by Benigni-Bossa lists tens of SAs for mutagenicity, and all of them
are also used to label carcinogenicity. This assumption may have not been adopted
by other models in the case of SAs for carcinogenicity. In the case of the statistical
models, the differences generally derive from three sources:
1. The chemicals at the basis of the model. There may be differences in their
number, their heterogenicity, and the nature of the chemicals. For instance, in the
216 E. Benfenati et al.

case of models for Ames mutagenicity, some models are based on about 20,000
substances [5], while others are based on much smaller collections [6]. This last
reference, for instance, relates to azo compounds, and this offers an example of
the difference related to the nature of the substances. It is obvious that the training
set is smaller for azo compounds. However, in some cases, it is useful to have
focused models for specific chemical classes. For instance, the Benigni-Bossa
model identifies the presence of the aromatic azo moiety and immediately assigns
as mutagenic the substance. Actually, as shown in Gadaleta et al. [6], from an
experimental point of view, half of the aromatic azo dyes are non-mutagenic.
This model was the basis of the effort to develop more accurate models, able to
provide better predictions for the specific chemical category. Thus, in the case of
azo compounds, the model, even if based on hundreds of substances, and not tens
of thousands, provides better results, because quite focused. The heterogenicity
of the training set is another obvious aspect. It is easier to obtain good models
if the set is homogeneous, but the applicability domain of the model is more
limited.
2. The chemical information. The way to address the chemical information may
provide very different results. This refers both to the format used as input and
to the descriptors used. In some cases, the ways to describe the molecule are
quite compact and simple. For some models, in particular old models, simple
physico-chemical information was used. Nowadays, there are thousands of chem-
ical descriptors and of fingerprints which are used to capture relevant features.
The format to represent the molecule can be simple, as the SMILES string, or bi-
or tri-dimensional representations. We have to mention that typically the highest
level of uncertainty and variability in toxicity models is associated to the exper-
imental values. Thus, in several cases, we found no advantages in the use of
tri-dimensional representation, versus bi-dimensional ones. Even SMILES are
sometimes used as direct input of the in silico models for toxicity, and this is a
convenient approach since there is no need of calculating molecular descriptors.
This is the case of the models based on CORAL [7, 8].
3. The algorithm used to build up the model is the third main component in the
model, which surely influences the model. Today there are more and more
models using advanced algorithms. Machine learning is largely applied, and
deep learning is used too. These recent algorithms surely provide novel solu-
tions and better possibilities to cope with large collections of data, and nonlinear
phenomena. Multitask modelling is also a recent interesting approach. The use
of these sophisticated algorithms is useful provided that there are enough data.
For smaller sets traditional methods are equivalent. These recent approaches
have been described recently [9, 10]. The multitude of tools also made clear that
there is not one single approach. Conversely, there are multiple ways to get quite
similar, equivalent results. This aspect implies a shift from the assumption and
the effort to identify the “perfect”, ideal model, because based on the proper
structural components and the correct equations at the basis of the biochemical
8 Computational Toxicological Aspects in Drug Design and Discovery … 217

process. Modelling toxicology is moving towards a probabilistic perspective, as

well as risk assessment in general [11].
There are other differences quite important when considering the in silico models
for toxicity predictions. Some models are regression models, i.e. provide quanti-
tative values as outcome, while others are classifiers. Most typically, for human
toxicity endpoints, the classifiers are binary ones, such as toxic or not. In some cases,
models have been developed to address the potency of the adverse level, such as
high, medium, or low. Quite often the models are indeed classifiers for most of the
endpoints listed at the beginning of this section. However, there are quantitative
models also for mutagenicity and carcinogenicity, taking into account the potency
[12, 13]. On the other hand, even if the lethal dose which kills 50% of the animal
(LD50) is a quantitative value, there are models which refer to threshold values
to classify substances as toxic or not, or with different levels of toxicity. In this
case, attention should be paid that the threshold values may differ depending on the
regulation and the country.
We mentioned above that there are commercial and free models, publicly available
on the internet. Of course, a clear difference is economic. Exercises which have been
done on different tools and endpoints did not identify differences in performance
between commercial and free models [14–17]. A main difference, beyond the price,
is the access level and the transparency and documentation. Free models have wide
access without restrictions. Quite typically the documentation is good, such as the
information on the substances in the training set and the algorithm. Some of these
tools, such as VEGA (www.vegahub.eu), are also open source. Conversely, typically
this kind of information is not available for commercial software. The algorithm
is proprietary, and the availability of the structures and toxicological data of the
substances within the training set may be difficult to obtain and report. Thus, the
commercial software is quite opaque. The documentation is requested within certain
regulations (like REACH [4]), and it is on the basis of the confidence in the results
obtained. The different models for the different endpoints certainly have a different
level of reliability depending on the endpoint. We discussed above the case of Ames
mutagenicity. This is an endpoint where the results are usually good. The property
is relatively simple, and there are in a few cases models based on about 20,000
substances, as we said. The general performance of the different models has been
compared and may vary depending on the kind of substances [16, 18, 19].
Other endpoints have different performance, and some large exercises have been
done [15–17]. The endpoints which refer to chronic toxicity or involve complex
toxicological processes are more difficult; this is the case of developmental toxicity
and reproductive toxicity. In this case, the data available are for a few hundreds
of compounds, and surely, the toxicological process is quite complex, involving in
some cases more than one generation. Great caution should be used to evaluate the
results of the models for these endpoints; multiple models should be used, and it is
recommended to carefully verify if there are similar substances with experimental
data supporting the final evaluation, according to a weight-of-evidence approach
which will be discussed below.
218 E. Benfenati et al.

Quite often recent in silico models do contain tools to evaluate if the prediction is
reliable. This is done by referring to the information present in the substances in the
training set and considering the so-called applicability domain. For instance, in the
case of the models in VEGA, the applicability domain is measured in a quantitative
way, and the software provides the applicability domain index (ADI) ranging from 0
to 1. This index is calculated by the software based on the chemical and toxicological
information, and on the algorithm [20]. Thus, in practice, the ADI looks at the most
similar compounds present in the set of substances at the basis of the model. This is a
first contribution of the ADI. This piece of information, which is purely chemical, is
addressed considering the similarity values of the most similar substances (see also
the following section regarding similarity), and the presence of unusual chemical
moieties. Another component of the ADI is specific for the property of the model,
such as toxicity. In this case, the software compares the predictions obtained on
the most similar compounds present in the set used to build the model. Of course,
the predictions are specific for the specific endpoint and are closely related to the
local situation represented by the most similar compounds. This value is called the
accuracy of predictions and is integrated within the ADI. Furthermore, the software
compares the predicted value of the target compound, with the experimental values
of the most similar compounds. This value is also very closely related to the property,
and thus it may change for different endpoints, even looking at the same substances.
Finally, for the ADI calculation, the software takes into account some factors related
to the algorithm, but these components usually have a lower impact on the final ADI.
Not all the in silico models have this complex approach as VEGA. In some cases,
the applicability domain is addressed only in a qualitative way, and the substance
is assigned only as inside or outside the applicability domain, as in the case of
the T.E.S.T. (https://www.epa.gov/chemical-research/toxicity-estimation-software-
tool-test) and Danish QSAR Database (https://qsar.food.dtu.dk/). Furthermore, the
applicability domain is typically calculated by comparing the target substance with
those in the training set, and this is done using tools for the chemical similarity but
not so often considering other factors related to the endpoint and the algorithm, as
done in VEGA. We will discuss more in detail this point later.

8.3 Tools for Read-Across

In silico models and read-across belong to the so-called non-testing methods. Read-
across has been used for decades by experts to evaluate substances. Even the process
of the identification of the SAs discussed above is somehow related to the concept
that some similar compounds present a common toxicological effect, simply because
they share some similar molecular moieties. The expert systems derived from this
strategy originate from a concept implicit in the read-across process. However,
exploring more cases, it has been found that there are substances which apparently
are similar but present different toxicological profiles. On the one hand, several rules
of exceptions to the SAs have been introduced; on the other hand, other features,
8 Computational Toxicological Aspects in Drug Design and Discovery … 219

not only based on the chemical similarities, have been introduced, to supplement
the approach with further inputs. Beyond the chemical similarity, toxicological
aspects (e.g. mode of action), physico-chemical properties, and toxicokinetic prop-
erties have been proposed. Experimental values and in silico predictions can be
used and assessed manually [21]. Here, we are more interested in the approaches
where computers have a higher role. In some cases, the read-across tools refer to
collections of data or programs which are heterogeneous, such as in vitro data or
information on metabolism, integrated with information based on the structure [22,
23]. In other cases, all the information necessary for the read-across derives from
the chemical structure, and the additional information related to the toxicological
aspects, for instance, is also derived from the chemicals structure, in the form of SA
or information on the mode of action [24, 25]. In this perspective, the advantage is
that the approach can aim to obtain the complete matrix of the values to be used
for read-across, solving a main critical aspect of traditional read-across: to strongly
rely on the data availability. Indeed, read-across is opportunistic and, traditionally,
depends on the experimental data. However, if we imagine using predictions, when-
ever experimental data are missing, we can potentiate the approach and have a more
reproducible strategy. Programs such as ToxRead and ToxDelta, both available within
the VEGAHUB platform (www.vegahub.eu), aim to address this aspect [24, 25].
Let us consider ToxRead as an example. It applies two processes of similarity
search: The first one is based on the structural similarity and the second one relates
to the specific property of interest, which can be a toxicological endpoint or another
endpoint. The structural similarity is calculated using the software developed for
the VEGA in silico models [20]. Particularly, this tool is based on a combination of
ways to represent and compare structures, related to the presence of certain compo-
nents, with the relative weights. These components and weights have been optimized
on millions of substances, and in general, the approach is quite robust. However, it
is important to comment a fundamental aspect of similarity. It is not an objective
property of a chemical, or in general of a certain item. Similarity always implies the
presence of at least two items, and it is related to the purpose of similarity. In practice,
in our case of substances and toxicity properties, for instance, the definition of simi-
larity is necessarily related to the endpoint. For instance, two substances may have a
similar fish bioconcentration factor value, but very different genotoxicity. Indeed, in
the case of the genotoxicity, there are peculiar SAs—i.e. peculiar fragments, which
may represent the occurrence of a toxicological process or not—but these fragments
may be neutral or not so relevant for bioconcentration. Let us consider the presence
of an epoxide versus an ether group. From the “point of view” of the bioconcentration
factor, the epoxy group is a kind of ether group, while regarding genotoxicity the
epoxy may imply a potential toxic effect, which is not observed in the case of the
ether group.
Based on this concept, ToxRead includes a second series of metrics for similarity
which are specific for a defined endpoint. Thus, in practice, ToxRead has tens of
modules, one for each endpoint. The tool for the chemical similarity is applied for
all modules, and then there are collections for rules, each collection specific for a
certain endpoint.
220 E. Benfenati et al.

For instance, in the case of Ames mutagenicity, ToxRead contains a collection of

more than 800 rules. These rules include the SAs of Benigni-Bossa, plus other collec-
tions, calculated with SARpy [26] or other algorithms, or also extracted manually.
An interesting observation is that the Benigni-Bossa SAs only represent fragments
associated with toxicity, while the other collections contain also fragments associated
with the lack of toxicity. In the case of properties with continuous values, ToxRead
has rules which are associated with threshold values.
Thus, ToxRead combines the two similarity measures, the structural one and
another related to the property. For the target compound, the user can visualize all
the SAs and rules associated with the effect. Thus, the software provides a general
view of the factors related to the presence and absence of effect. As we said, ToxRead
contains both kinds of fragments, pointing towards effect or lack of effect. This fact
is an improvement compared to the expert-based approach, which is quite subjective,
and based on personal experience. Conversely, ToxRead is moving towards a more
systematic approach, which is more objective and reproducible, compared to the
“manual” read-across.
It is also important to notice that if A is similar to B and B is similar to C,
we cannot conclude that A is similar to C. In practical terms, this means that the
similarity application is quite local, and it loses its utility moving away from the very
similar substances. There are multiple algorithms to measure similarity, and in many
cases, the similarity is normalized between 1 and 0. The different metrics do not
overlap; thus, the comparison should be done internally, within each software. For
instance, in the case of the structural similarity within the VEGA tools (VEGA in
silico models and ToxRead), 1 means identity, and substances should be considered
with good similarity if the similarity is above 0.9 or 0.85. If the similarity is lower
than 0.75, the two substances contain important dissimilar parts, but these values are
not “official”, unique thresholds; conversely, there is a tendency, and these values
are only indicative and vary by the endpoint and the substance. Similarity value
is a key factor for read-across, but the number of similar substances is also very
important. If the similar substance is only one, this implies uncertainty. If there are
more similar substances with quite close property values, this is much better. Read-
across is very sensitive to noise, indeed, and if it is based on a single substance, the
quality of the value of the source substance should be high (in read-across we call
target substance the substance to be evaluated, and source substances the substances
with the experimental values used for read-across). This is a limitation associated
with read-across compared to the in silico models. In silico models are using multiple
substances, as we have seen, even tens of thousands, and thus in this case if there
is noise, and substances with data of lower reliability, this is not a critical aspect. In
the case of read-across, we must be quite sure about the data quality of the source
substances, because for read-across, very few substances are typically used and in
some cases even one.
If we use more than one substance, which is preferable, we should use interpolation
preferably, avoiding extrapolation: thus, if we have a set of substances with a carbon
chain of different lengths, we should have substances with the lengths longer and
8 Computational Toxicological Aspects in Drug Design and Discovery … 221

shorter than the chain of the target substance. Compared to in silico models, read-
across may also have advantages. If we have very similar compounds, the overall
assessment may be more robust if based on read-across compared to the in silico
models. The in silico models relate to the global population; thus, the read-across
may be more robust in a local situation. We will discuss this point later in more
details.
Another important point related to read-across is that the results are dependent on
the similarity metrics, associated thresholds, and number of similar substances that
we use. For instance, ToxRead allows selecting the number of substances for read-
across. If all the similar substances have the same label (in the case of classifiers) or
close property values (in the case of continuous values), we can derive our conclusion
quite easily. Conversely, if the similar compounds in the cluster for read-across are
not homogeneous, regarding the property value, we are in a critical situation. This
will be discussed later addressing weight-of-evidence.

8.4 Weight-of-Evidence

In the case of non-testing methods, and in general considering experimental values, it

is common to use multiple values; in several cases, the values may derive from hetero-
geneous sources. This represents an issue, regarding the process of comparing and
integrating multiple values. The European Food Safety Authority (EFSA) addressed
this within a specific guidance document [27]. This guidance indicates that the user
should proceed sequentially (1) gathering all data, (2) evaluating the data separately,
and then (3) integrating the results of the process. The process of the evaluation is
detailed in the guidance, and basically, for the integration the user should evaluate
these three aspects: (a) the relevance, (b) the reliability, and (c) the consistency of
the multiple data.
The relevance should be evaluated regarding the specific purpose, thus with refer-
ence to the problem formulation. This aspect is quite important for instance in the
case of read-across. Let us imagine we have a source compound, which is mutagenic
and contains a SA related to mutagenicity. In this case, we should check if the SA is
also present in the target compound. If the SA is not present in the target compound,
this similar compound is not relevant in our case. Indeed, the mutagenicity is due,
very probably, to the SA, but if this SA is not present in the target substance, this
information is irrelevant. Conversely, if the SA is present in both the source and in
the target compounds, this source is surely relevant. At this point, we can investigate
the reliability of the information regarding the SA. To do this, we can check if the
similar substances with the SA are mutagenic or not (we already commented that
there is a certain number of substances with the SA which are not active, depending
on the specific SA, and there may be rules of exception for a certain SA). Thus, if we
observe that similar substances are not active, we can conclude that for the specific
case the reliability of the SA is low.
222 E. Benfenati et al.

Similarly, in the case of in silico models, it is often possible to evaluate the

reliability of the result since several models do provide measurements regarding this.
We have already discussed above that models quite often apply tools to evaluate the
applicability domain, and we said that in some cases this measurement is a continuous
value, while in other cases, this is addressed as inside or outside. Of course, in the
case of a quantitative value, as for the VEGA models, we have a refined appreciation
of the reliability domain. If the value is quantitative, it is also possible to assign a
weight to the individual in silico models, and then integrate the results by applying
different weights for each model. Thus, in the case of an individual in silico models,
considering the reliability, one approach is to use both the value of the prediction
and its reliability by applying a weight, and this is used for instance in the case
of the VEGA consensus model for mutagenicity where four individual models are
combined according to the scheme described in the literature [28]. Other platforms
of in silico models integrate the results of the individual models in a different way,
because they cope with the applicability domain in a categorical way, as inside
or outside the applicability domain. This is the case of the T.E.S.T. software of
the US EPA (https://www.epa.gov/chemical-research/toxicity-estimation-software-
tool-test), and the Danish QSAR Database (https://qsar.food.dtu.dk/). These systems
accept or not the models based on threshold values for the applicability domain, and
then all the accepted models are considered equivalent regarding their reliability.
There are different possibilities to integrate the values from in silico models and
read-across, and we discussed them [29].
Once the relevance and the reliability of each line of evidence have been char-
acterized, the last point, as we discussed above, is to integrate the separate lines of
evidence. Referring to the EFSA guidance above mentioned [27], and to the iden-
tification of separate lines of evidence to be integrated, we notice that the in silico
models (at least some of them) provide three lines of evidence:
1. The prediction. This is the value given by the model, which is supported by the
descriptors, algorithms, etc.
2. Similar compounds. They are shown by some models, such as VEGA and
T.E.S.T., for instance. This line of evidence should be used as read-across, as
discussed above.
3. The potential mechanism involved in the process. This of course is a piece of
information which may not be present. This depends on the model, the substance,
and the endpoint. Indeed, some models do not contain the indication about
the mechanism, because they are not built using this piece of information. For
instance, a model based on the kNN algorithm (in which the prediction is based
on the k most similar compounds of the training set, combined usually by a mean
or a median) ignores the mechanism.
Not all the in silico models are so descriptive and detailed. The VEGA format, as
an example, is quite rich regarding these pieces of information. Thus, the user should
analyse these three lines of evidence separately and then compare them. In case of
conflicting lines of evidence, it is useful to refer to the process we discussed above
regarding the relevance and the reliability of each line of evidence. For instance,
8 Computational Toxicological Aspects in Drug Design and Discovery … 223

we had the example above of a similar substance which was not relevant, because
it contained a SA that is not present in the target compound. In this case, the read-
across based on this substance should be disregarded. Conversely, it may be that
the read-across indicates the presence of a very similar compound which is toxic,
while the prediction is for non-toxicity. In this case, the high similarity implies high
relevance, and thus, this line of evidence prevails unless we have a clear explanation
why the similar compound is toxic, for a certain reason, which does not apply to the
target substance.

8.5 Tools for Integrating Multiple Endpoints

In [30], the European Commission indicated a general strategy to reach a toxic-free

environment, minimizing and substituting the substances of concern and promoting
the development of chemicals sustainable by design. This means that assessors have
to identify the riskiest substances. Typically, this is done by analysing the properties
that lead considering a chemical as a substance of very high concern (SVHC). How
identify an SVHC is defined by laws or regulations. Depending on the regulation
of reference (even among the European regulations), the thresholds may differ [31].
In this paragraph, we will refer to the REACH [4] definition that considers SVHC
the chemicals that are persistent, bioaccumulative and toxic (PBT), very persistent
and very bioaccumulative (vPvB), carcinogenic, mutagenic, or reprotoxic (CMR),
or endocrine disruptors (ED).
From a computational point of view, the identification of the SVHC means the
integration of various evidence for several endpoints. It is a further step in the inte-
gration of the available information (see the previous paragraph). The assessor has to
integrate the information available for each endpoint (e.g. for the endpoint persistence
in water, he/she has to integrate half-life in water, ready biodegradability information,
hydrolysis, etc.), several endpoints to evaluate a property (e.g. persistence in water,
sediment and soil to evaluate the persistence), and then combine all the properties
(e.g. persistence, bioaccumulation and toxicity for the PBT/vPvB assessment).
Tools like VEGA can process several chemicals and several endpoints at the same
time, but the results are not integrated; the user has to do it manually or using other
tools. Moreover, different users may obtain different integrated results, depending on
the integration strategy used. For this reason, in the last years, several methods and
tools to assess the PBT/vPvB, the CMR, and/or the ED properties were developed. We
can divide them into two categories, the screening, and the prioritization (or ranking)
tools. The screening tools divide the list of chemicals into two (e.g. toxic and non-
toxic) or a few classes (e.g. toxic, moderately toxic, non-toxic). The prioritization
tools assign a score to each chemical, which allows to order them from the most to the
less concerning. Both can be useful, depending on the purpose. Industries may want
to evaluate several possible substances in an early design phase, before synthesizing
them, to decide which ones can proceed in the development process. In this case, a
screening tool may be sufficient. If a regulatory body wants to decide on the most
224 E. Benfenati et al.

concerning substances to plan management strategies and has to focus its attention
on a subgroup of the concerning chemicals (e.g. the most concerning ones), it needs
a prioritization tool.
Here we present, as an example, the JANUS tool (https://www.vegahub.eu/),
developed to prioritize chemicals based on seven properties—i.e. persistence (P),
bioaccumulation (B), toxicity (T), carcinogenicity (C), mutagenicity (M), repro-
toxicity (R), and ED—and the REACH [4] thresholds. It responds to the specific
requirements of the German Umweltbundesamt (UBA) to rank the registered chem-
icals from the most to the less hazardous. This means that a screening approach,
like the one proposed in [32], is not sufficient. With JANUS, the user has a tool
that runs QSAR models for more than 20 endpoints, integrates these predictions in
a sort of automatic weight-of-evidence with the experimental values, if available, to
assess the seven properties, and integrates them into three scores for the prioritization
always considering the reliability of the values. In this way, two chemicals with the
same assessment value but with different reliabilities will have different prioritiza-
tion scores. Moreover, it offers the possibility to evaluate the microbial metabolites
in the same way as the parental.
More in detail, JANUS considers firstly the seven properties separately. The prop-
erties are evaluated considering the presence of experimental values; they can be
inserted by the user or retrieved by the VEGA models implemented in JANUS.
Indeed, JANUS runs 48 VEGA models that can be used as key predictions or as
supporting information (e.g. to modulate the reliability depending on their concor-
dance with the key prediction). In some cases, like P and B, there are screening
classes. They are classes of substances (e.g. perfluoroalkylic compounds) recognized
as hazardous (e.g. P) but not well predicted by the models. For these chemicals, an
arbitrary assessment and reliability are assigned. In the case of multiple values with
the same reliability, their agreement is also considered (i.e. the disagreement reduces
the reliability of the property).
The output of the first part is one assessment value and its reliability that are
combined into property scores. The property scores are then integrated into three
different prioritization scores. The first one is based only on the P and B properties
(with equal weight), the second is based on the human health-related properties
(C, M, R and ED, combined on a worst-case approach), and the last is based on
all the properties (combining the human health properties and T on a worst-case
approach). The scores range from 0 (non-hazardous) to 1 (hazardous). They will
be close to 1 for hazardous chemicals with good reliability and close to 0 for non-
hazardous chemicals with good reliability. The chemicals with scores around 0.5
may be moderately hazardous with good reliability or chemicals (hazardous or not)
with low reliability.
As mentioned above, an advantage of the JANUS tool is the possibility to
run the metabolism module. It is based on the public EAWAG Biocatalysis/
Biodegradation Database (https://envipath.org) and generates the metabolites of the
first step of microbial biodegradation. The metabolites are then processed as the
parental compound to allow the user to analyse the possible concerns derived from
the metabolites.
8 Computational Toxicological Aspects in Drug Design and Discovery … 225

Literature reports several other tools or methods for screening and prioritization.
Some of them are summarized below.
. Strempel et al. [33] report a screening method for PBT and vPvB properties
based on predicted and experimental values. The output is four classes—PBT,
nonPBT0 (none of the properties is of concern), nonPBT1 (one of the properties
is of concern), and nonPBT2 (two properties are of concern). For each property,
the value is divided by the assigned threshold to obtain a property score. The PBT
and the vPvB score are the average of the property scores.
. Böhnhardt [32] reports a strategy based on predicted log Kow and biodegradation.
The authors, applying a screening approach, identified 132 chemicals to be deeply
analysed (starting from a list of 4445 chemicals).
. Another approach (https://repository.qsartoolbox.org/Tools/List/Profilers) was
developed by the European Chemicals Agency (ECHA); it is based on three
screening profilers (one for each of the three properties, P, B and T) that use
predicted and experimental values applying a workflow through the OECD QSAR
toolbox (https://qsartoolbox.org/). They allow classifying chemicals as persistent
or not, bioaccumulative or not, or toxic or not (considering ecotoxicity only),
respectively.
. In [34], the authors developed a tool to identify the SVHC based on chemical
similarity. This screening method was tested on an external set in [35] and became
a freely available tool [36].
. Carlsen and Walker [37] proposed a ranking based on partial order theory that
uses as input predicted values for P, B, and T.
. Shin et al. [38] developed a scoring system based on exposure and hazard
indicators, collected from databases, to rank chemicals for the occupational
environment, therefore, only for human health risk.
. Papa and Gramatica [39] proposed the PBT index, a screening tool for the PBT
assessment based on a multiple-linear regression equation that uses four descrip-
tors. It considers the cumulative PBT behaviour. The revised version is freely
available (http://www.qsar.it/). The output is an index with a threshold to classify
chemicals as PBT or non-PBT. The authors defined it as a precautionary approach
[40].

8.6 Tools for Integrating Hazard and Exposure

Benefit-risk assessment is the core task for marketing applications for new drugs and
to decision making throughout the life cycle of any medicinal product [41]. The ever-
increasing inclination to place safe products on the market has led to an evolution
of the methodologies adopted to assess drugs but also other consumer products. The
change was sealed by the next generation risk assessment (NGRA), an approach
characterized by decision making without the use of animal testing [42], in line with
the paradigm shift induced by the new European Cosmetic Regulation [43, 44].
226 E. Benfenati et al.

Within the European LIFE VERMEER project (LIFE16 ENV/IT/00016)

(https://www.life-vermeer.eu/), an innovative strategy to integrate hazard and expo-
sure assessment for human and environmental risks was designed, with the ambitious
goal to harmonize and facilitate the risk assessment process in Europe, increasing
human and environmental health prevention. A battery of new software was devel-
oped and made freely available to the scientific community worldwide. These tools
were developed ad hoc for specific case studies, such as cosmetics, food contact
materials (FCM), solvents, biocides, dispersants, and oil fractions. However, it is
important to highlight that the architecture of these tools is reproducible and adapt-
able to drugs, or for instance to medical devices, which could be target categories
for the after-life plan foreseen within the project. Indeed, despite the project ended
in April 2022, it was thought in a future-oriented way, with flexible capabilities for
future improvements in mind.
For the chemical assessment, a great number of models and tools exist, some
of them freely available, others commercial. Individual and separate models for
hazard and exposure assessment must be run in order to perform a risk assessment.
Moreover, the application and the interpretation of these models could be intricate,
making the integration of various information really challenging [44]. Within the
LIFE VERMEER project, new comprehensive and holistic systems were built, inte-
grating models for hazard and exposure within the same platform, offering an inno-
vative and forward-looking solution that can substantially increase the perspective
in the field of risk assessment. These novel systems represent an inducement to the
use of new approach methodologies, often belittled, or at least not fully included in
the daily routine of risk assessors.
As previously indicated, the software we have developed is focused on specific
commercial categories. Thus, they were designed following the regulatory frame-
work, including information retrieved from European legislations and guidance
drafted by authorities, such as the European Chemical Agency (ECHA), the European
Food Safety Authority (EFSA), or the Scientific Committee on Consumer Safety
(SCCS) and containing specific thresholds or conditions of use. This represents
another key point, which distinguishes these new tools from those already available
on the market. Moreover, these new software were developed trying to get inside
the user’s mind, replicating with an “in silico” design, the expert system approach.
For specific sectors, such as cosmetics, where animal testing was banned, in silico
methodologies represent the new frontier for risk assessment and VERMEER tools
fit perfectly in this context, providing a ground-breaking solution to assess products.
VERMEER tools are freely downloadable from the VERMEER website (https://
www.life-vermeer.eu/) and VEGAHUB (https://www.vegahub.eu/).
For the cosmetics case study, we have developed VERMEER Cosmolife, an inno-
vative tool for the risk assessment of cosmetic products, which represents, for this
sector, the first prototype ever able to integrate within the same platform the two
pillars of risk assessment, hazard, and exposure [44]. VERMEER Cosmolife repli-
cates the risk assessment procedure followed by regulators, covering the four main
steps of risk assessment (hazard identification, exposure assessment, dose–response
assessment, and risk characterization) [45]. Therefore, the expert system approach is
8 Computational Toxicological Aspects in Drug Design and Discovery … 227

partially embedded in the framework of the tool; moreover, several statistical-based

tools were incorporated into the structure. The tool, indeed, has some QSAR models
to predict mutagenicity, genotoxicity, skin sensitization, and no observed adverse
effect level (NOAEL), as well as a tool for the threshold of toxicological concern
(TTC) [44].
The software allows evaluating the toxicological profile of cosmetics ingredi-
ents, providing at the same time a well-defined indication of exposure scenarios
related to specific product types in order to characterize risk for consumers. For
the exposure, the calculation provided by the SCCS Notes of Guidance 11th revi-
sion [45] was adopted, with a refinement based on new models for skin permeation.
VERMEER Cosmolife was built considering the regulatory framework for cosmetics;
in particular, it complies with the requirements of the European Regulation 1223/2009
[43]. An important aspect of this system is that it manages simultaneously multiple
ingredients enabling a comprehensive evaluation of a typical cosmetic formulation
and projecting the attention to real and practical applications. Moreover, the tool is
extremely user-friendly, helping, even more, the end user. Finally, the tool has been
designed with flexible capabilities for future extension. Additional features will be
added in future, taking into account new approaches and different lines of evidence,
by exploiting other in-house models and tools, already described in the previous
paragraphs. Some examples of how the tool can be used are present in the work of
Selvestrel et al. [44].
Whereas VERMEER Cosmolife is a Java stand-alone application, the other
VERMEER tools developed within the LIFE VERMEER project were implemented
within MERLIN-Expo, a platform for simulating the fate of a chemical in the envi-
ronment and the human body [46] (https://merlin-expo.eu/). Some VEGA models
have been included into MERLIN-Expo in order to create an integrated tool for
risk assessment. One of these “MERLIN-Expo-based” tools is VERMEER FCM
which provides information with respect to exposure (i.e. migration) and hazard
endpoints for chemicals (e.g. additives, etc.) intended to be used in plastics food
contact materials. VERMEER FCM was developed taking into consideration the
European Regulation 10/2011 [47] and the EFSA Notes for Guidance [48].
With this software, it is possible to predict the concentration of chemical migrants
in food in contact with FCM. The predicted concentrations depend on several param-
eters such as the contact time between food and FCM, the contact temperature, the
material type (e.g. type of plastic polymer) as well as important physico-chemical
properties of the chemical migrants (e.g. lipophilicity). According to the EFSA Notes
for Guidance, toxicological requirements depend on the migration of the chemical
into the food. The tool allows predicting various toxicological endpoints required
for regulatory purposes. Among them: in vitro mutagenicity, in vitro micronucleus
formation, sub-chronic oral toxicity, carcinogenicity, and developmental toxicity.
The tool allows running both deterministic and probabilistic simulations, the last
of them based on the Monte Carlo algorithm. The tool is freely available on the
VERMEER (https://www.life-vermeer.eu/) and VEGAHUB (https://www.vegahu
b.eu/) websites.
228 E. Benfenati et al.

Shifting the attention to the environmental sphere, let us start to consider the
VERMEER Rodenticides tool. The VERMEER Rodenticides tool is used to provide
exposure and hazard assessments regarding the release of rodenticides in surface
waters. This tool is able to predict the concentration of rodenticides in aquatic organ-
isms considering at the same time ecotoxicological endpoints. Also in this case,
the regulatory framework plays a fundamental role. Rodenticides are considered as
biocidal products and are then submitted to the related European regulation (Regu-
lation (EU) No 528/2012) [49]. In order to facilitate the evaluation of environmental
risks associated with rodenticides, ECHA [50] has defined a set of generic scenarios,
i.e. a set of conditions about sources, pathways and use patterns of active compounds.
One of these scenarios concerns the application of rodenticides on bank slopes of
watercourses like rivers, drainage channels, lakes, ponds, lagoons, etc. However,
rodenticides can be flushed away due to high rainfall directly into surface waters.
The structure of VERMEER Rodenticides is very similar to the FCM one with the
difference that other kinds of parameters and physico-chemical properties are needed
for the simulation. As in the case of VERMEER FCM, VERMEER Rodenticides
allows running both deterministic and probabilistic simulations.
Another sector considered within the VERMEER project is that of solvents. In
the synthesis of active pharmaceutical ingredients, solvents play a crucial role and, in
the context of the green chemistry, the choice of sustainable and “greener” solvents
is essential to preserve the environmental impact [51]. According to this assumption,
a new tool for the environmental risk assessment of solvents was developed, and
it is freely available. The identification of green solvents is demanding because a
lot of parameters (health effect, environmental impact, physico-chemical properties,
etc.) have to be taken into account [52]. The VERMEER Solvents tool represents a
first attempt to build an innovative system for the assessment of solvents, useful also
for pharmaceutical industries, focused on this first prototype, on the environmental
health. New features including other critical aspects, previously mentioned, will
be implemented in future versions of the tool. The structure reflects that of the
VERMEER Rodenticides tool.
Finally, a new tool called VERMEER Dispersants was developed, which allows
simulating the distribution of oil components within an aquatic ecosystem under
different environmental conditions. It allows focusing on the comparison of the
distribution of components with or without the addition of chemical surfactants
that support the dispersion of the oil. Oils spills can severely impact the marine
environment; therefore, measures to reduce potential damages shall be taken. The
application of dispersants, which boost the transformation of floating oil into small
droplets, is a valid and effective option to at least reduce this problem [53]. With
VERMEER Dispersants, it is possible to predict the concentration of oil components
in different environmental compartments of the marine ecosystem over time. Even
in this case, both deterministic and probabilistic simulations can be run.
These tools represent an impressive achievement in the field of risk assess-
ment because they move forward on the real case application and because they
have a forward-looking fingerprint. Some of these new tools are already used
by industries, and they are representing the starting point for new projects,
8 Computational Toxicological Aspects in Drug Design and Discovery … 229

such as SILIFOOD (https://www.sciensano.be/en/projects/applicability-silico-tools-

support-risk-assessment-non-evaluated-substances-migrating-food-contact) funded
by Belgian Authorities and FANGHI, funded by the Lombardy Region in
Italy (https://www.openinnovation.regione.lombardia.it/it/b/38399/forme-avanzate-
di-gestione-dei-fanghi-di-depurazione-in-un-hub-innovat). Moreover, from these
outcomes, new tools will take shape for other case studies.

8.7 Innovation and Caution in Safe-by-Design Drug

Production

At the heart of Safe-by-Design is the idea that when innovating a production process,
all risks related to a target product should be as much as possible anticipated. In such
an attempt, the precautionary principle plays a central role. According to The Rio
Declaration on Environment and Development (1992) [54], “lack of full scientific
certainty shall not be used as a reason for postponing cost-effective measures to
prevent environmental degradation”. The practical implementation of a Safe-by-
Design approach, however, does also call for innovative methods, so that the innova-
tion principle is on an equal footing with the precautionary principle [55]. But what
should be the appropriate balance between the two principles when pharmaceutical
production is considered? In this case, it is possible to tackle the issue from a double
perspective. The first concerns the safety of the patient, who is supposed to take the
drug for a specific medical reason. The second covers the environmental and societal
risks involved in the manufacturing process through which the drug is synthesized.
From the medical perspective, innovation could play a positive role. Looking at the
most promising information technology advances, artificial intelligence techniques
are expected to lead to a deeper critical assessment of protein structure prediction
(CASP) in the early steps of drug design and discovery. Campos et al. [56] proposed a
concept of molecular editing capable of insertion, removal, or modification of atoms
in extremely functionalized chemicals at will and in a precise fashion with computa-
tional tools’ involvement. The same authors have shown how analogues of a complex
lead scaffold might be edited via heteroaromatic reduction, site-selective C−H func-
tionalization, ring contraction, or ring expansion, evading a hypothetically lengthy
synthesis of analogues followed by synthetic hurdles. Integrated chemical databases
with these Web servers like the use of comparative toxicogenomic database (CTD)
(http://ctdbase.org/) for human health may lead to anticipate potential toxicolog-
ical problems. Notwithstanding, the cautionary principles underlying medical ethic
generally leads to recommend the use of old, rather than new and innovative drugs.
In general, from the medical perspective, we observe how innovation is often treated
with a legitimate suspicion. Particularly, knowledge concerning drug toxicological
profiles is regarded well consolidated only after years of monitoring its post-market
adverse effects, so a new drug is taken into account only after more conventional
230 E. Benfenati et al.

remedies failed to solve the clinical problem. These criteria can also operate in non-
pharmaceutical contexts. Likewise medical protocols are often applied to discourage
the use of novel drugs whenever it is possible; the approach that we implemented
in ToxEraser to select a cosmetic targeted to a specific functional use, does also
encourage the substitution of an ingredient on the grounds of the evidence consoli-
dated through the systematic assessments of authoritative and regulatory institutions
(https://www.life-vermeer.eu/). On the other hand, when the attention is turned to
the manufacturing processes, innovation looks rather universally welcome. These
processes are more and more blamed for the environmental risk to which society is
exposed, as they emerge to be some of the less safe and sustainable of all industrial
processes. The medical and regulatory requirements of pharmaceutical purity are the
main reasons leading to more waste per kilogram product as compared to making less
sophisticated compounds of less stringent purity [55]. To some extent, this problem
may be related to the emphasis given to the patient’s safety, that is, just the first of
the two addressed perspectives. The idea of green chemistry and its principles has
been known since 1990, but the real implementation of rules in drug designing and
synthesis is still limited. In our experience, medicinal chemists are generally inter-
ested in contemplating environmental parameters in pharmaceutical production, but
they did not understand the parameters that play a role in the “greenness” of a
molecule, the kind of assays that exist or may be developed that measure environ-
mentally relevant parameters, or how such parameters could be used in compound
selection or lead optimization. Among the parameters of interest to control envi-
ronmental risks, we may find the drug persistence in the environment, either due to
lack of biodegradation processes or to the mobility of the drug. Further endpoints of
eco-toxicological can also be informative, particularly those depending on membrane
permeability. Other useful screening can be focused on P450, kinase assays, and reac-
tive metabolites, and the bioaccumulation potential may represent worthwhile targets
of assessment. In turn, by impacting on mobility via solubility, permeability, absorp-
tion via modification of chemical binding forces, bioaccumulation, even a phisco-
chemical property as simple as the lipophilicity may inform on multiple endpoints
of environmental interest. Though in the beginning, the improvement of methods of
synthesis and purification of a targeted molecule is under scrutiny for an expanding
class of pharmaceutical substances. By looking at a molecule of interest as the final
output of a multi-step process, the attention is first focused on the manufacturing steps
which involve the highest environmental and societal risks. Several indicators can be
used for this purpose. Some of them are based on the ratio of the total mass of waste
and the mass of the final product (E factor), which considers all wastes (reagents,
solvent losses, aids, and fuel) included in the process, excluding the final product and
water, which are generally not considered under waste. Furthermore, the conversion
efficiency of a chemical process in terms of all atoms involved and the desired prod-
ucts (Atom Economy factor) was also advocated to anticipate the potential impact on
the environment during drug production [51]. All analyses and investigations point
at the use of solvents as the major detrimental environmental factor and hazard in
the workplace. Solvents are volatile organic compounds employed in large volumes
and leading to high waste, pollution, and health hazards. The concern did lead to
8 Computational Toxicological Aspects in Drug Design and Discovery … 231

several classification systems that can be adopted for practical reference, like the
ICH Q3 Class, the Concentration Limit in Pharmaceuticals, and Ranking of Major
Organic Solvents in the Form of Hazardous Impact [52]. The use of as few solvents
as possible is certainly a guiding principle. Water may be more often employed for
the same purposes; other dangerous solvents are deemed necessary at a first super-
ficial glance. However, when the use of water cannot be an option, we find that
the criteria underlying the “greenness” metric of the classification of solvents cover
many parameters, like occupational health, quality (risk and amount of impurities),
utilization of complete reagent, recyclability issues, risk of residual solvents in the
pharmaceutical product and final cost. To support the choice of solvents within such a
complex multidimensional framework, several tools are made available for designing
the green synthesis process under the US EPA and Green Chemistry Expert System
(GCES), which offers the GC process guidelines and support information as well
as the green solvents/reaction conditions module [57]. More generally, the selection
of appropriate solvents, reaction conditions, substrates, and types of reactions can
be checked for millions of compounds by permutation and combination, utilizing in
silico approaches even before practical synthesis within a period of an hour. This is
likely to offer much greater and broader options to synthetic chemists [51].

8.8 Tools for Building New Models

Above we presented many tools which can be used for different purposes (hazard
identification, risk assessment, prioritization, etc.), related to different endpoints.
However, there are many endpoints which are covered by any model, not with
standing the user would have an interesting collection data about a set of substances to
deveolp a new, specific model. There are several tools which can be easily applied to
develop new models, without a high level of experience. The tools listed within the
VEGAHUB site represent just an example (in the section download, https://www.veg
ahub.eu/download/), but there are more tools available. A great advantage of many of
these tools is that they can be downloaded and thus used internally, by the industry.
Frequently, industries have collections of data which are proprietary, of restricted
use, to be exploited within the research and development phase. Conversely, other
tools, such as those discussed above, have dual access, since ideally the same system
for the toxicity assessment should be used by industry and regulators.

8.8.1 aiQSAR

This tool, which contains a set of pre-built models, can be also used to develop new
models, both as classifier and regression models. The system builds up a battery of
local models, obtained with a few tens for substances which are selected based on
232 E. Benfenati et al.

the similarity of the substance to be predicted. Thus, in a certain way, the system is
something between the read-across and the in silico models.

8.8.2 DTC LAB Tools

Developed by the team led by Prof Roy at the Jadavpur University in India, they offer
a series of chemometric tools to build in silico models. This system includes models
for regression, classification, read-across, as well as specific tools for mixtures and
nanomaterials.

8.8.3 SARpy

Within VEGA some of the implemented models have been developed using SARpy,
but this tool can be used to develop classifiers for the desired endpoint. The software
uses the SMILES format. It starts fragmenting the molecule into fragments which
are smaller and smaller. Each fragment is associated with an effect label, based on its
prevalence in active or inactive substances. It involves a sequential process, since the
substances already predicted by the model are eliminated, and then new fragments
are searched.

8.8.4 QSARpy

QSARpy allows developing models for continuous endpoints using the chemicals of
the training set and a list of fragments, named modulators, to which a quantitative
value (positive, negative or null) is assigned. Modulators are extracted by fragmenting
the structures and calculating the difference between the couples of chemicals of the
training set that differs only by the fragment. If subtracting and/or adding one or
more modulators to a chemical of the training set is possible to obtain the target, the
prediction is done by subtracting and/or adding the value assigned to the modulators
to the chemical of the training set.

8.8.5 CORAL

CORAL uses the SMILES of the molecule to build up models. It identifies if a

certain combination of characters, present in the SMILES, are associated with the
effect. These combinations of characters contain few symbols, thus in practice do
not represent large fragments. CORAL is quite versatile and can incorporate other
8 Computational Toxicological Aspects in Drug Design and Discovery … 233

features, provided by the user, which are not the classical chemical information.
For example, the user can specify the size of the nanomaterial, the temperature, the
duration of the experiment, etc.

8.8.6 SOM Tool

The National Institute of Chemistry in Slovenia developed this tool based on a

counter-propagation neural network. Also in this case, some of the models developed
in that institute with this methodology have been implemented in VEGA, but the user
can develop his/her model using neural networks with this software.

8.8.7 OCHEM

Within the EC project CONCERT REACH (https://www.life-concertreach.eu/), a

network of four systems has been established: VEGA, OCHEM, the Danish QSAR
Database, and AMBIT. OCHEM is a software system offering many tools to develop
in silico models, also applying some of the most recent approaches, such as deep
learning.

8.8.8 AMBIT

AMBIT has been developed with the support of the CEFIC, the European Council
of the Chemical Industries. This system, related to collections of chemicals derived
from the registered substances in Europe, is quite powerful for read-across. It offers
powerful systems for data exploration and takes into account the real substances,
which means the substance with its components.

8.9 Conclusions

We need efficient, fast, and powerful tools to explore the potentially adverse effect
of pharmaceuticals. The process of the development of new pharmaceuticals is very
expensive and involves a set of sequential steps. We analysed above the in silico
tools which can be applied to investigate adverse effects. We have seen that there
are multiple tools and multiple purposes. The trend is towards a more and more
systematic and pre-organized way to handle the information at the basis of the adverse
properties. The sequential scheme, with steps to be done at successive times, is based
on the fact that in the past these steps were done through laboratory experiments on
234 E. Benfenati et al.

real substances, involving multiple tests. Each of these tests has a cost, and it was
convenient to find an optimal scheme. The use of the in silico model represents
a novelty in this. Initially, some in silico models have been used to mimic some
experimental tests. This is a possibility, but we have shown that the use of the in
silico models may have a more advantageous impact, with dramatic influences on
the scheme, which may be consistently modified. To run one or tens of models does
not make a lot of difference for the computer. The number of endpoints, and the
number of substances, would no more represent a barrier. This opens completely
new perspectives. We can anticipate effects on a large palette of features.
Furthermore, the possibility to run tools in parallel implies a better possibility
to identify links between properties and features, and the computer is able to cope
with complexity in a better way than humans. We have seen that there are tools to
integrate tens of models on different endpoints and platforms for exposure and hazard
at the same time. New tools offer ways to identify safer substances, addressing both
the adverse properties and the functional use. This is indeed the frontier. In silico
models can help to organize these multiple, heterogeneous data. This process should
be organized within the same architecture. However, part of the components have
to be public, because necessary to evaluate the risk—and thus, both industry and
regulators need access—and part of the components should be restricted. These are
the tools to explore the beneficial, functional use of the substances.

Acknowledgements We thank the EC, LIFE programme for the LIFE CONCERT REACH project
(LIFE 17 GIE/IT/000461).

References

1. Benfenati E (ed) (2022) In silico methods for predicting drug toxicity, 2nd edn. Springer, New
York
2. EMA (2018) ICH M7 assessment and control of DNA reactive (mutagenic) impurities in
pharmaceuticals to limit potential carcinogenic risk
3. ICH Harmonised Tripartite Guideline (2017) Assessment and control of DNA reactive
(mutagenic) impurities in pharmaceuticals to limit potential carcinogenic risk—M7
4. European Parliament, Council of the European Union (2006) REGULATION (EC) No 1907/
2006 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of 18 December 2006
concerning the Registration, Evaluation, Authorisation and restriction of Chemicals (REACH),
establishing a European Chemicals Agency, amending Directive 1999/45/EC and repealing
Council Regulation (EEC) No 793/93 and Commission Regulation (EC) No 1488/94 as well
as Council Directive 76/769/EEC and Commission Directives 91/155/EEC, 93/67/EEC, 93/
105/EC and 2000/21/EC
5. Gini G, Zanoli F, Gamba A et al (2019) Could deep learning in neural networks improve
the QSAR models? SAR QSAR Environ Res 30:617–642. https://doi.org/10.1080/1062936X.
2019.1650827
6. Gadaleta D, Porta N, Vrontaki E et al (2017) Integrating computational methods to predict
mutagenicity of aromatic azo compounds. J Environ Sci Health C Environ Carcinog Ecotoxicol
Rev 35:239–257. https://doi.org/10.1080/10590501.2017.1391521
8 Computational Toxicological Aspects in Drug Design and Discovery … 235

7. Toropova AP, Toropov AA, Marzo M et al (2018) The application of new HARD-descriptor
available from the CORAL software to building up NOAEL models. Food Chem Toxicol
112:544–550. https://doi.org/10.1016/j.fct.2017.03.060
8. Toropov AA, Toropova AP, Selvestrel G et al (2020) Prediction of no observed adverse effect
concentration for inhalation toxicity using Monte Carlo approach. SAR QSAR Environ Res
31:1–12. https://doi.org/10.1080/1062936X.2020.1841827
9. Gini G, Benfenati E (2021) From data to models. In: Chemometrics and cheminformatics in
aquatic toxicology. Wiley, pp 89–124
10. Gini G (2022) QSAR Methods. In: Benfenati E (ed) In silico methods for predicting drug
toxicity. Springer, US, New York, NY, pp 1–26
11. Maertens A, Golden E, Luechtefeld TH et al (2022) Probabilistic risk assessment—the keystone
for the future of toxicology. ALTEX 39:3–29. https://doi.org/10.14573/altex.2201081
12. Toropov AA, Toropova AP, Benfenati E (2009) QSAR modelling for mutagenic potency of
heteroaromatic amines by optimal SMILES-based descriptors. Chem Biol Drug Des 73:301–
312. https://doi.org/10.1111/j.1747-0285.2009.00778.x
13. Toma C, Manganaro A, Raitano G et al (2021) QSAR models for human carcinogenicity: an
assessment based on oral and inhalation slope factors. Molecules 26:127. https://doi.org/10.
3390/molecules26010127
14. Honma M, Kitazawa A, Cayley A et al (2019) Improvement of quantitative structure–activity
relationship (QSAR) tools for predicting Ames mutagenicity: outcomes of the Ames/QSAR
international challenge project. Mutagenesis 34:3–16. https://doi.org/10.1093/mutage/gey031
15. Mansouri K, Kleinstreuer N, Abdelaziz AM et al (2020) CoMPARA: collaborative modeling
project for androgen receptor activity. Environ Health Perspect 128:027002. https://doi.org/10.
1289/EHP5580
16. Mansouri K, Abdelaziz A, Rybacka A et al (2016) CERAPP: collaborative estrogen receptor
activity prediction project. Environ Health Perspect 124:1023–1033. https://doi.org/10.1289/
ehp.1510267
17. Mansouri K, Karmaus AL, Fitzpatrick J et al (2021) CATMoS: collaborative acute toxicity
modeling suite. Environ Health Perspect 129:047013. https://doi.org/10.1289/EHP8495
18. Benfenati E, Golbamaki A, Raitano G et al (2018) A large comparison of integrated SAR/
QSAR models of the Ames test for mutagenicity$. SAR QSAR Environ Res 29:591–611.
https://doi.org/10.1080/1062936X.2018.1497702
19. Van Bossuyt M, Van Hoeck E, Raitano G et al (2018) Performance of In silico models for
mutagenicity prediction of food contact materials. Toxicol Sci 163:632–638. https://doi.org/
10.1093/toxsci/kfy057
20. Floris M, Manganaro A, Nicolotti O et al (2014) A generalizable definition of chemical
similarity for read-across. J Cheminformatics 6:39. https://doi.org/10.1186/s13321-014-0039-1
21. Van der Stel W, Carta G, Eakins J et al (2021) New approach methods supporting read-across:
two neurotoxicity AOP-based IATA case studies. Altern Anim Experimentation : ALTEX
38:615–635. https://doi.org/10.14573/altex.2103051
22. Gadaleta D, Bakhtyari AG, Lavado GJ et al (2020) Automated integration of structural, biolog-
ical and metabolic similarities to improve read-across. ALTEX 37:469–481. https://doi.org/10.
14573/altex.2002281
23. Helman G, Shah I, Williams AJ et al (2019) Generalised read-across (GenRA): a workflow
implemented into the EPA CompTox Chemicals Dashboard. ALTEX 36:462–465. https://doi.
org/10.14573/altex.1811292
24. Gini G, Franchi AM, Manganaro A et al (2014) ToxRead: a tool to assist in read across and its
use to assess mutagenicity of chemicals. SAR QSAR Environ Res 25:999–1011. https://doi.
org/10.1080/1062936X.2014.976267
25. Golbamaki A, Franchi AM, Manganelli S et al (2017) ToxDelta: a new program to assess how
dissimilarity affects the effect of chemical substances. Drug Des 06. https://doi.org/10.4172/
2169-0138.1000153
26. Ferrari T, Cattaneo D, Gini G et al (2013) Automatic knowledge extraction from chemical
structures: the case of mutagenicity prediction. SAR QSAR Environ Res 24:365–383. https://
doi.org/10.1080/1062936X.2013.773376
236 E. Benfenati et al.

27. Committee ES, Hardy A, Benford D et al (2017) Guidance on the use of the weight of evidence
approach in scientific assessments. EFSA J 15:e04971. https://doi.org/10.2903/j.efsa.2017.
4971
28. Cassano A, Raitano G, Mombelli E et al (2014) Evaluation of QSAR models for the prediction
of Ames genotoxicity: a retrospective exercise on the chemical substances registered under
the EU REACH regulation. J Environ Sci Health C 32:273–298. https://doi.org/10.1080/105
90501.2014.938955
29. Benfenati E, Chaudhry Q, Gini G, Dorne JL (2019) Integrating in silico models and read-across
methods for predicting toxicity of chemicals: a step-wise strategy. Environ Int 131:105060.
https://doi.org/10.1016/j.envint.2019.105060
30. COM (2020) Communication from the commission to the European parliament, the council,
the European economic and social committee and the committee of the regions. https://ec.eur
opa.eu/environment/pdf/chemicals/2020/10/Strategy.pdf
31. Moermond CTA, Janssen MPM, de Knecht JA et al (2012) PBT assessment using the revised
annex XIII of REACH: a comparison with other regulatory frameworks. Integr Environ Assess
Manag 8:359–371. https://doi.org/10.1002/ieam.1248
32. Böhnhardt A (2013) Identification of potential PBT/vPvB-Substances by QSAR methods.
Federal Environment Agency (Germany)
33. Strempel S, Scheringer M, Ng CA, Hungerbühler K (2012) Screening for PBT chemicals
among the “Existing” and “New” chemicals of the EU. Environ Sci Technol 46:5680–5687.
https://doi.org/10.1021/es3002713
34. Wassenaar PNH, Rorije E, Janssen NMH et al (2019) Chemical similarity to identify potential
substances of very high concern—an effective screening method. Comput Toxicol 12:100110.
https://doi.org/10.1016/j.comtox.2019.100110
35. Wassenaar PNH, Rorije E, Vijver MG, Peijnenburg WJGM (2021) Evaluating chemical simi-
larity as a measure to identify potential substances of very high concern. Regul Toxicol
Pharmacol 119:104834. https://doi.org/10.1016/j.yrtph.2020.104834
36. Wassenaar PNH, Rorije E, Vijver MG, Peijnenburg WJGM (2022) ZZS similarity tool: the
online tool for similarity screening to identify chemicals of potential concern. J Comput Chem
43:1042–1052. https://doi.org/10.1002/jcc.26859
37. Carlsen L, Walker J (2003) QSARs for prioritizing PBT substances to promote pollution
prevention. QSAR Comb Sci 22:49–57
38. Shin S, Moon H-I, Lee KS et al (2014) A chemical risk ranking and scoring method for the
selection of harmful substances to be specially controlled in occupational environments. Int J
Environ Res Public Health 11:12001–12014. https://doi.org/10.3390/ijerph111112001
39. Papa E, Gramatica P (2010) QSPR as a support for the EU REACH regulation and rational
design of environmentally safer chemicals: PBT identification from molecular structure. Green
Chem 12:836. https://doi.org/10.1039/b923843c
40. Gramatica P, Cassani S, Sangion A (2015) PBT assessment and prioritization by PBT Index
and consensus modeling: comparison of screening results from structural models. Environ Int
77:25–34. https://doi.org/10.1016/j.envint.2014.12.012
41. Davies M, Lane S, Shakir SF (2020) Principles of benefit-risk assessment: a focus on some
practical applications. In: FPM. https://www.fpm.org.uk/blog/principles-of-benefit-risk-assess
ment-a-focus-on-some-practical-applications/
42. Dent MP, Vaillancourt E, Thomas RS et al (2021) Paving the way for application of next
generation risk assessment to safety decision-making for cosmetic ingredients. Regul Toxicol
Pharmacol 125:105026. https://doi.org/10.1016/j.yrtph.2021.105026
43. European Commission EC (2009) Regulation (EC) No.1223/2009 of the European parliament
and of the council of 30 November 2009 on cosmetic products. Official J Eur Union L 342:59–
209
44. Selvestrel G, Robino F, Baderna D et al (2021) SpheraCosmolife: a new tool for the risk
assessment of cosmetic products. ALTEX 38:565–579. https://doi.org/10.14573/altex.2010221
45. SCCS—Scientific Committee on Consumer Safety (2021) SCCS notes of guidance for the
testing of cosmetic ingredients and their safety evaluation—11th revision
8 Computational Toxicological Aspects in Drug Design and Discovery … 237

46. Ciffroy P, Alfonso B, Altenpohl A et al (2016) Modelling the exposure to chemicals for
risk assessment: a comprehensive library of multimedia and PBPK models for integration,
prediction, uncertainty and sensitivity analysis—the MERLIN-Expo tool. Sci Total Environ
568:770–784. https://doi.org/10.1016/j.scitotenv.2016.03.191
47. EC—European Commission (2011) Commission Regulation (EU) No 10/2011 of 14 January
2011 on plastic materials and articles intended to come into contact with food Text with EEA
relevance
48. EFSA Panel on Food Contact Materials, Enzymes, Flavourings and Processing Aids (CEF),
Silano V, Bolognesi C et al (2008) Note for Guidance for the preparation of an application for
the safety assessment of a substance to be used in plastic food contact materials. EFSA J 6:21r.
https://doi.org/10.2903/j.efsa.2008.21r
49. EC—European Commission (2012) Regulation (EU) No 528/2012 of the European parliament
and of the council of 22 May 2012 concerning the making available on the market and use of
biocidal products. Official J Eur Union L 167, 1–123
50. European Chemicals Agency (2018) Revised emission scenario document for product type
14: rodenticides. Available on https://echa.europa.eu/documents/10162/16908203/esd_pt14_
en.pdf/d27d3b7e-9aa6-8146-9228-f464901b526e. Publications Office, LU
51. Kar S, Sanderson H, Roy K et al (2022) Green chemistry in the synthesis of pharmaceuticals.
Chem Rev 122:3637–3710. https://doi.org/10.1021/acs.chemrev.1c00631
52. Prat D, Hayler J, Wells A (2014) A survey of solvent selection guides. Green Chem 16:4546–
4551. https://doi.org/10.1039/C4GC01149J
53. Grote M, van Bernem C, Böhme B et al (2018) The potential for dispersant use as a maritime
oil spill response measure in German waters. Mar Pollut Bull 129:623–632. https://doi.org/10.
1016/j.marpolbul.2017.10.050
54. United Nations (1992) 1992 Rio declaration on environment and development—Centre for
international law
55. Cue BW, Zhang J (2009) Green process chemistry in the pharmaceutical industry. Green Chem
Lett Rev 2:193–211. https://doi.org/10.1080/17518250903258150
56. Campos KR, Coleman PJ, Alvarez JC et al (2019) The importance of synthetic chemistry in
the pharmaceutical industry. Science 363:eaat0805. https://doi.org/10.1126/science.aat0805
57. EPA Green Chemistry GCES Tool. In: American Chemical Society. https://www.acs.org/
content/acs/en/greenchemistry/research-innovation/tools-for-green-chemistry.html. Accessed
1 Mar 2021
Chapter 9
Read-Across and RASAR Tools
from the DTC Laboratory

Arkaprava Banerjee and Kunal Roy

Abstract In silico approaches for activity/toxicity predictions have gained attention

recently, and these are accepted by various regulations like EU-REACH. Aspects like
reproducibility, less ethical complications, no animal use and reduced time are some
of the reasons why researchers nowadays are shifting toward the in silico approaches
for prediction. Quantitative Structure–Activity Relationship (QSAR) is one of the
most commonly used in silico approaches for the prediction of response, but the
only drawback is that since it involves model-derived predictions, it is prone to erro-
neous results when the number of training data points is insufficient. In recent times,
similarity-based algorithms like Read-Across are being adopted by researchers with
the aim of data gap filling. The Read-Across approach does not involve model-
derived predictions, rather it involves similarity-based predictions and thus can effi-
ciently be used for data gap filling. The authors at the DTC Laboratory have devel-
oped a Java-based Read-Across tool (https://sites.google.com/jadavpuruniversity.in/
dtc-lab-software/home) which utilizes three different similarity-based approaches
(Euclidean Distance-based, Gaussian Kernel Similarity-based and Laplacian Kernel
Similarity-based) for the prediction of responses of the query compounds along with
the external validation metrics and the overall error measures. Moreover, the compu-
tation of certain compound-specific similarity and error-based metrics enables the
user to identify the uncertainty in the Read-Across-based predictions, especially
when the observed response values of the query compounds are unreported. The
idea of clubbing the QSAR methodology and the Read-Across approach together
has given rise to a novel chemometric prediction approach termed as Read-Across
Structure–Activity Relationship (RASAR). The authors at the DTC Laboratory are
the pioneers in reporting the quantitative predictions using the RASAR approach (q-
RASAR). A Java-based RASAR descriptor calculator tool has also been developed
which calculates the similarity and error-based descriptors based on the similarity-
based approach selected by the user. The authors feel that these tools have a lot

A. Banerjee · K. Roy (B)

Department of Pharmaceutical Technology, Drug Theoretics and Cheminformatics (DTC)
Laboratory, Jadavpur University, Kolkata 700032, India
e-mail: [email protected]
URL: http://sites.google.com/site/kunalroyindia/

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 239
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_9
240 A. Banerjee and K. Roy

of potential in bridging data gaps and may prove to be very much essential for
the predictions of various property/activity/toxicity endpoints in the future.

Keywords Read-across · RASAR · Tools · DTC laboratory

9.1 Introduction

In the context of risk assessment and environmental safety, chemical compounds

are regulated by different legislations in the European Union (EU) like registration,
evaluation, authorization and restriction of chemicals (REACH) (EC regulation No.
1907/2006) and classification, labeling and packaging of substances and mixtures
(CLP) (EC regulation 1272/2008) in addition to different application-specific pieces
of legislation for cosmetic, plant protection and biocidal products and legislation
addressing food, novel food and food contact materials [1]. Although toxicity testing
exercises have traditionally been performed in experimental animal-based studies,
in the recent past, there has been increasing focus on the sustainability of these
methodologies [2]. Reliable toxicity analysis methods to identify, assess and inter-
pret the deleterious properties of any substance are urgently needed. To avoid the
ethical complications and minimize animal use, the replacement, refinement and
reduction in animal experimentations (3R principles) of Russell and Burch can be
used [3]. There is strong support from regulatory bodies like US Environmental
Protection Agency (US EPA), European Chemical Agency (ECHA), Organiza-
tion for Economic Co-operation and Development (OECD), etc. for the develop-
ment of New Approach Methodologies (NAMs) that meet regulatory preparedness
[4]. NAMs include development novel omics, in vitro and computational methods
including modeling and Read-Across, searchable databases that can be used for
grouping and Read-Across purposes, computational modeling of quantitative struc-
ture–activity relationships, dose–response assessments and modeling, analyses of
biological processes and toxicity pathways, etc. [5]. Computational toxicology helps
to identify hazards of compounds even before synthesis, and thus, they help in predic-
tions in very early stages of drug development. Computational methods can aid in
gap-filling and guide risk minimization strategies in the chemical industries and also
in regulatory settings. While there is a need to develop robust and reliable non-animal
methods, no single alternative method is expected to provide a unique replacement
for assays targeted for more complex toxicological endpoints. Hence, results from
a combination of techniques including computational modeling, in vitro assays,
high-throughput screening, omics and mathematical biology can provide comple-
mentary information to develop a comprehensive picture of the potential response
of an organism to a chemical substance. Adverse outcome pathways (AOPs) of
stressor chemicals and systems biology frameworks enable logical integration of
relevant information from diverse sources [6]. Computational data include results
obtained from quantitative structure–activity relationship (QSAR) models, chem-
ical categories, grouping, Read-Across and physiologically-based pharmacokinetic
9 Read-Across and RASAR Tools from the DTC Laboratory 241

(PBPK) models and “big data” analysis [7]. There are general and more specific
factors to be considered when using different computational methods as described
in the multitude of guidance documents available to support their use for regulatory
purposes. While computational methods are currently mainly used more for internal
rather than regulatory decision-making, the situation may change as confidence grows
in their applicability and predictivity. It is advisable to use computational methods
within a weight of evidence (WoE) approach and with all available data. In one
hand, computational models are valuable cheap alternatives to in vitro and in vivo
experiments, and on the other, their use by non-experts can eventually be misleading
[8].
Read-Across is a non-testing data gap-filling technique that provides informa-
tion for toxicological hazard potential based on the known toxicity data of source
compound(s) with a “similar” property or chemical profile [9]. Read-Across, i.e.,
the local similarity-based intrapolation of properties, is gaining importance with
increasing data availability and guidelines on how to process and report it. It is mainly
applied to in vivo test data as a gap-filling approach, but can as well be used for other
incomplete datasets [10]. Molecular similarity provides a simple and popular method
for virtual screening of chemical databases. Molecular diversity analysis explores the
coverage of a given structural space and underlies many approaches for compound
selection and design of combinatorial libraries. In chemoinformatics, molecular simi-
larity and diversity measures are complementary. The measures of molecular simi-
larity and/or diversity involve in general three main components: descriptors, their
coefficients and the weighting scheme [11]. The increased usage of Read-Across
was driven by the huge expenditure with respect to money, time and manpower apart
from ethical issues associated with in vivo testing and also encouraged by regula-
tory frameworks (like EU-REACH) in order to minimize animal experimentation.
ECHA and OECD have published several guidelines on the technicalities of a Read-
Across study. Read-Across is an evolving method with several open issues as well as
opportunities. As per the ECHA’s Read-Across assessment framework, the starting
point is chemical similarity [12]. There are several approaches and algorithms avail-
able for calculating chemical similarity based on molecular descriptors, fingerprints,
distance/similarity measures and weighting scheme for specific endpoints. Toxico-
logical endpoints are usually in the focus of Read-Across cases. In order to further
enhance the quality of Read-Across cases, new approach methods can be very useful.
While computational models offer major benefits to regulators and toxicologists, the
absence of a guidance document for the execution of computational experiments and
use of the results in an integrated framework may lead to uncertainty and contra-
dictions across models and users, even for the same chemicals. Read-Across offers
a strategy for deriving reference points or points of departure for risk assessment
of untested chemicals, from the available experimental data for structurally similar
compounds, mostly based on expert judgment [13]. While drug toxicity pathways
can be extremely complex and difficult to fully understand [14], specific parts of the
pathway may be simpler to understand. Every toxicity pathway starts with a molec-
ular initiating event (MIE), which if well understood makes it possible to predict
242 A. Banerjee and K. Roy

which compounds can be involved in that particular MIE with the help of compu-
tational techniques. Structural alerts can be used to identify chemicals which can
form a covalent bond with a biological macromolecule. Prediction of the toxicity
of a compound requires a comparison with similar compounds causing the same
MIE and that are associated with known toxicological data. It is possible to form
categories of compounds that are all thought to act via the same MIE and then use
Read-Across within the category to make a toxicity prediction.
Enoch et al. [15] presented a mechanistic Read-Across for predicting the skin
sensitization potential of alkenes acting via Michael addition using the elec-
trophilicity index as a measure of similarity for sensitizing chemicals. The index
was shown to offer a chemically interpretable qualitative ranking of the chemicals
within the Michael acceptor domain. Schuurmann et al. [16] developed a Read-
Across method based on atom-centered fragments (ACFs) for evaluating chemical
similarity for predicting fish toxicity. The study showed that increasing the ACF
minimum similarity increases the prediction quality while decreasing the applica-
tion range. Kuhne et al. [17] presented a Read-Across approach that makes use of the
atom-centered fragment (ACF) method as quantitative measure for structural simi-
larity for quantitative prediction of the acute toxicity of organic compounds toward
the water flea Daphnia magna. Hartung [10] presented a new web-based tool called
REACH-across, which aims to support and automate structure-based Read-Across.
Russo et al. [18] discussed identification and integration of biological data from
various resources and used the in vitro bioassay data-driven profiling strategy for
Read-Across modeling.
Although traditional Read-Across approaches are based on the chemical simi-
larity principle to predict chemical toxicity, complexity in the mechanism of biolog-
ical activity and/or toxicity makes the accuracy of such predictions often inadequate
justifying the usage of biological similarity in addition to chemical similarity for
Read-Across predictions. Low et al. [19] developed a hazard classification and visu-
alization method using both chemical structural similarity and biological response
similarity measured in multiple short-term assays. The Chemical−Biological Read-
Across (CBRA) approach determines each compound’s toxicity from both chemical
and biological analogues whose similarities are determined by a similarity coeffi-
cient like Tanimoto coefficient. Ravenzwaay et al. [20] suggested that metabolomics
can be used for chemical grouping and Read-Across from a biological perspective
which can reduce animal testing and provide with mechanistic interpretation of the
biological action. Przybylak et al. [21] stressed the importance of consideration of
biotransformation to metabolites having the same mechanism of electrophilic reac-
tivity, via the same metabolic pathway, with a rate of transformation sufficient to
induce the same in vivo outcome for the rat oral repeated-dose toxicity of β-olefinic
alcohols.
Schultz et al. [22] have identified a variety of uncertainties including the regulatory
use of the prediction, the data for the endpoint being assessed, the Read-Across argu-
mentation and the similarity justification that can potentially impact acceptance of a
Read-Across argument. Alves et al. [23] introduced the multidescriptor Read-Across
(MuDRA) method which is conceptually related to the well-known kNN approach
9 Read-Across and RASAR Tools from the DTC Laboratory 243

using different types of chemical descriptors simultaneously for similarity assess-

ment. They found that models derived from the MuDRA approach show high predic-
tion Accuracy similar to that of conventional QSAR models. The authors claimed
MuDRA to provide a powerful alternative to a much more complex consensus QSAR
modeling.
Luechtefeld et al. [24] recently combined the chemical similarity concept (Read-
Across) with supervised learning methods resulting in a new technique termed as
Read-Across structure–activity relationship (RASAR). They used binary fingerprints
and Jaccard distance to define chemical similarity. A large chemical similarity adja-
cency matrix was constructed from which feature vectors were derived for supervised
learning. A “simple” RASAR trains a logistic regression model to predict chemical
hazards based on the similarity to the closest chemical which has tested positive
(maxPos) and similarity to the closest chemical tested negative (maxNeg). The “Data
Fusion” (DF) RASAR extends this concept by expanding the feature vectors using
all available property data rather than only the modeled endpoint. This version of
RASAR trains random forest models from diverse chemical information of analogs.
Wu et al. [25] used standard properties of chemicals along with similarity measures
in a DF-RASAR approach and showed efficient predictions of chemical hazards
across taxa. They showed that DF-RASAR has several advantages in the inte-
gration of the data from different effects. AbdulHameed et al. [26] developed a
chemical-similarity-based protocol for the prediction of the potential of a chemical
to interact with different toxicity targets. They evaluated the performance of 2D
and 3D similarity approaches in correctly ranking known interacting compounds
using an external evaluation set from the ChEMBL database. They found that the 2D
similarity-based predictions were superior to the 3D approaches.
This chapter summarizes the Read-Across and RASAR tools and different quality
and evaluation metrics associated with this research developed in the Drug Theoretics
and Cheminformatics (DTC) Laboratory and their applications in the prediction of
different activity/toxicity endpoints.

9.2 The Theory Behind the Read-Across Approach

The concept of Read-Across utilizes the similarity between compounds to predict the
response values of the query compounds. This technique has recently emerged as one
of the most promising techniques for data gap filling, especially in cases where there
is shortage of experimental data [27]. As supported by the Organization for Economic
Co-operation and Development (OECD), the Read-Across approach can efficiently
replace in vivo testing. If other techniques like High-Throughput Screening can be
coupled with Read-Across, it can enhance the quality of predictions. The Regis-
tration, Evaluation, Authorization and Restriction of Chemicals (REACH) in the
European countries have preferred experimentally unavailable toxicity data gener-
ated using in silico approaches [1]. Also, in the European Union (EU), there is a
244 A. Banerjee and K. Roy

ban on animal experimentation for the evaluation of cosmetics, and such evalua-
tion should be carried out using alternative and in silico approaches. To fulfill the
requirements of REACH, the Read-Across approach needs to fulfill certain criteria:
• The results obtained should be sufficient to perform risk assessment, classification
and labeling.
• There should be enough coverage of the key aspects in the testing methods.
• The duration of exposure should be comparably longer than the corresponding
test method, if this parameter has sufficient relevance.
• There should be reliable and adequate documentation of the applied method.
Read-Across approach can be further classified based on the differences in the
number of source and target compounds used for predictions. The four different
strategies of Read-Across predictions are as follows:
• One-to-One: This approach makes use of the similarity of a single-source
compound to predict the response value of a single-target compound.
• One-to-many: In this approach, there is utilization of a single-source compound to
predict the response values of multiple-target compounds, based on the similarity
levels.
• Many-to-one: This approach involves two or more source compounds to predict
the response of a single-target compound, based on the similarity levels.
• Many-to-many: Involvement of two or more source compounds to predict the
response values of multiple-target compounds based on similarity levels.
Although Read-Across is a very useful technique to fill data gaps, there are a
couple of problems which can be encountered. First, due to the non-availability of
sufficient proof supporting justification, it is difficult to ascertain the absence of
toxicity from the Read-Across predictions. Secondly, it does not give us the idea
about the uncertainty in predictions [27]. The first issue can be addressed by linking
the Read-Across technique with Molecular Initiating Events and Adverse Outcome
Pathways, thus giving rise to a concept called Biological Read-Across. The second
issue has already been addressed in our Read-Across tool by computing similarity
and error-based measures (vide infra) for each of the compounds, which enables the
user to assess the uncertainty in predictions.

9.3 Read-Across Tool from the Drug Theoretics

and Cheminformatics Laboratory

The Read-Across tool, developed by the DTC Lab, is a Java-based program which
quickly computes Read-Across-based quantitative predictions of endpoints and their
corresponding external validation metrics in terms of Q 2F1 and Q 2F2 (correlation-
based metrics) [28]. It also computes overall error measures in terms of Root Mean
Squares Error of Predictions (RMSEP) and Mean Absolute Error (MAE). Even
9 Read-Across and RASAR Tools from the DTC Laboratory 245

without the availability of the observed endpoint values of the target compounds,
this tool can efficiently predict the possible endpoint values, but the external vali-
dation metrics cannot be computed as they require the observed response values
of the target compounds for their computation. This tool reaches one step ahead,
as it also calculates various compound-specific error measures for the individual
target compounds with respect to its nearest source compounds. With the help of
this tool, it is also possible to perform classification-based Read-Across, and the
input response does not necessarily need to be graded, i.e., it can handle quantitative
response values even while performing classification-based Read-Across. The tool
generates five output files—one of them displaying the close target compounds for
each query compound with their response values and similarity levels in a sorted
manner, while the second one is the main output file for Read-Across predictions
and their quantitative validation metrics. The rest three files show classification-based
validation metrics like Sensitivity, Specificity, Accuracy, Precision, Matthew’s Corre-
lation Coefficient (MCC), etc. The Receiver Operating Characteristic Curve (ROC
Curve) is also generated by taking each response value of the target compound as
the threshold and calculating the corresponding true-positive rate (Sensitivity) and
false-positive rate (1-Specificity) along with computation of the Area Under the
Curve (AUC). With the help of this tool, one can also proceed for quantitative Read-
Across Structure–Activity Relationship (RASAR) (vide infra) [29] by utilizing the
error measures for the individual target compounds as descriptors. This tool has
been upgraded to incorporate such features, the current version of which is avail-
able as Read-Across-v4.1 from https://sites.google.com/jadavpuruniversity.in/dtc-
lab-software/home. Figure 9.1 demonstrates the detailed workflow followed by the
Read-Across tool.
This tool utilizes three distance/similarity-based approaches, namely Euclidean
Distance-based, Gaussian Kernel Similarity-based and Laplacian Kernel Similarity-
based, for computation of the predicted response values and validation with the
external validation metrics. The descriptors are standardized at first to scale their
range, which ultimately reduces noise during distance/similarity calculations. The
Euclidean Distance approach computes the scaled Euclidean Distances between the
query compound and the source compounds using the following equation:
√
d= ∑(q − p)2 . (9.1)

In Eq. (9.1), d stands for the Euclidean Distance, q and p are the descriptor vectors
of the source and target compounds, respectively. It is important to know that the
Euclidean Distance approach does not involve any hyperparameter, and thus, the
optimization can only be performed with respect to the close source compounds and
distance threshold values. The Gaussian Kernel Similarity method is derived from
the Euclidean Distance method, and it computes the similarities between the query
and the source compounds. Mathematically, the Gaussian Kernel Similarity can be
represented as:
246 A. Banerjee and K. Roy

Fig. 9.1 Workflow of the Read-Across tool

||X i −Yi ||2

f (GK) = e− 2σ 2 . (9.2)

In Eq. (9.2), f (GK) stands for the Gaussian Kernel Similarity value, ||X – Y||2 is
the L 2 norm or square of the Euclidean Distance, and σ is a hyperparameter which
can be optimized. Thus, in the case of Gaussian Kernel Similarity-based predictions,
the hyperparameter σ, the number of close source compounds and the similarity
threshold are the optimizable entities, unlike Euclidean Distance-based approach,
which allows only the optimization of the number of close source compounds and
the distance threshold. The third similarity approach of predictions which this tool
provides is the Laplacian Kernel Similarity method. Unlike Gaussian Kernel Simi-
larity approach, this method utilizes the Manhattan Distance for the estimation of
the similarity between a particular source compound and the target compounds. The
mathematical representation for calculation of the Laplacian Kernel Similarity is as
follows:

f (LK) = e(−γ ||X −Y ||1 ) . (9.3)

Equation (9.3) demonstrates the equation for calculation of the Laplacian Kernel
Similarity. f (LK) is the Laplacian Kernel Similarity value, while ||X – Y||1 is the
Manhattan Distance between the source and the target compounds. Here, γ is the
hyperparameter, which allows the optimization of the Laplacian Kernel-based Read-
Across predictions.
9 Read-Across and RASAR Tools from the DTC Laboratory 247

9.3.1 Pre-requisites for Using This Tool

9.3.1.1 System Specifications

Read-Across-v4.1 does not require exhaustive system resources, i.e., it can run on
computers having standard memories for RAM and HDD/SSD. However, since it
is a Java-based software tool, it is necessary that a particular system needs to have
Java installed before running this tool. The Java Development Kit (JDK) can be
downloaded from https://www.oracle.com/java/technologies/downloads/, and after
successful installation, the Read-Across tool can be executed.

9.3.1.2 Input File Specifications

The program asks the user to enter two files, namely the training and test sets. The
input files should be a Microsoft Excel workbook, having the extension .xlsx. The
data tabulated in each of the training and test set files should have the following
specific pattern:
1st column: compound number.
2nd to nth column: descriptors in subsequent columns.
(n + 1)th column [last column]: biological activity/property/toxicity.
It is essential to note that the program can handle both quantitative and graded
response values (i.e., 0 and 1) depending upon the requirement of the user. In case
the input values are graded (0 and 1), the quantitative validation metrics should be
ignored.
Figure 9.2 shows snapshots of the sample training and test set input files.
This tool provides another option for the user to calculate only the biological
activity/property/toxicity of the target compounds without evaluating the quality of
predictions. To implement this feature, the user only needs to put “999” in the first
observed response value of the test set and any random entry for other compounds.

Fig. 9.2 Snapshots of the sample training and test set input files
248 A. Banerjee and K. Roy

In this case, the quantitative validation metrics and classification–based metrics are
not computed, and the ROC Curves are not generated.

9.3.2 Downloading and Execution of the Software

• The software in the form of a .zip file has been made available at the DTC lab tools
supplementary webpage (https://sites.google.com/jadavpuruniversity.in/dtc-lab-
software/home).
• The.zip file needs to be downloaded and the contents need to be extracted. One
can find that it consists of a folder (Read-Across-v4.1) inside which there is a .jar
file, a library folder, and two sample input files (Fig. 9.3a).
• The training and test set files need to be placed inside this Read-Across-v4.1
folder, i.e., the same folder which contains the.jar file (Fig. 9.3b).
• By double clicking on the Read-Across-v4.1.jar file, the program will be
executed (Fig. 9.3c).
• Certain dialog boxes which ask for appropriate data appear on the screen. The
user needs to enter the file names for the training and test sets, some constant
values (sigma and gamma, suggested value being 1), the number of close training
set compounds (which can range from 2 to 10, provided that they lie inside the
specified distance/similarity threshold), threshold values for distance (suggested
value being 0.5; 1 in case of no known threshold) and similarity (suggested value
in the range of 0–0.05; 0 in case of no known threshold). For classification, the
user needs to enter the file name for the test set and the threshold value, if the input
response values are quantitative (for example, the mean value of the responses of
the source data set and 0.5 if there are graded response values as input) (Fig. 9.4).
• Sorting of the similarity measures will be automatically printed in a newly gener-
ated file, namely TestSetFileName_Sort.xlsx, and the biological activities along
with the validation metrics will be automatically printed in a newly generated
file, namely TestSetFileName_Biological Activity.xlsx. Additionally, three other
files are also generated, namely TestSetFileName_Euclidean.xlsx, TestSetFile-
Name_Gaussian.xlsx and TestSetFileName_Laplacean.xlsx, which contain the
values of different classification metrics and the generated ROC Curves. The user
can see all these files generated in the same folder (Read-Across-v4.1) (Fig. 9.5).

9.3.3 Analysis of the Output Files

This program generates a total of five different output files, each of which encodes
certain chemometric information. The TEST_Biological Activity.xlsx file contains
the predicted response values of the query compound(s), their external validation
metrics in terms of Q 2F1 and Q 2F2 and the overall error measures in terms of Root Mean
Squares Error of Predictions (RMSEP) and Mean Absolute Error (MAE). Apart from
9 Read-Across and RASAR Tools from the DTC Laboratory 249

Fig. 9.3 Snapshots a after extraction of the .zip file, b after placing the training and test set files in
the folder, c the executable.jar file
250 A. Banerjee and K. Roy

Fig. 9.4 Snapshots during execution of the tool

9 Read-Across and RASAR Tools from the DTC Laboratory 251

Fig. 9.5 Snapshot of the generated output files

that, various other compound-specific error measures are also generated which help
the user to estimate the uncertainty of predictions for a compound with unreported
response values. The metric SD_Activity computes the weighted standard deviation
of the observed response values of the “n” close training compounds. SE stands
for Standard Error of the activity values of the “n” close training compounds. CV_
Activity represents the coefficient of variation of the observed response values of the
close “n” source compounds, while CV_Similarity represents the coefficient of varia-
tion of their similarity values. The metric MaxPos represents the maximum similarity
value of the target compound with respect to the closest source compounds having
response values greater than the threshold (training set response mean). Similarly,
MaxNeg denotes the maximum similarity value of the target compound with respect
to the closest source compounds having a response value lower than the threshold
(training set response mean). The metric g is a concordance measure, which takes
into account the positive fraction (fraction of compounds among the close source
compounds which have observed response values greater than the threshold), and uses
it to estimate the uncertainty of Read-Across-based predictions. The mathematical
representation of g is as follows:

g = 1 − 2|Posfrac − 0.5|. (9.4)

The metric average similarity defines the average similarity values among the
close “n” source compounds. SD_Similarity denotes the standard deviation of the
similarity values of the close “n” source compounds. gm [Banerjee-Roy coefficient]
is the modified version of g, which takes into account both the MaxPos or MaxNeg
and the Positive Fraction, and uses it to estimate the uncertainty in predictions. This
252 A. Banerjee and K. Roy

was developed to establish a directionality to distinguish between the probable active

and inactive compounds (with respect to the threshold observed response values).
gm can be mathematically represented as:

gm = (−1)n 2|Posfrac − 0.5|. (9.5)

The value of n is 1, when the value of MaxPos < MaxNeg, and n = 2 when MaxPos
> MaxNeg.
The TEST_Sort.xlsx file contains the sorted similarity values of the source
compounds along with their response values for each query compound, calcu-
lated with the three different similarity-based algorithms (Euclidean Distance-based,
Gaussian Kernel Similarity-based and Laplacian Kernel Similarity-based).
The TEST_Euclidean.xlsx, TEST_Gaussian.xlsx and TEST_Laplacian.xlsx files
compute the true-positive and false-positive rates, i.e., Sensitivity and 1-Specificity
values, taking each response value as the threshold and an ROC curve is computed
for each of the three similarity-based measures. Apart from this, various other
classification-based validation metrics are generated which determines the quality
of the predictions while performing classification-based Read-Across.

9.3.4 Application of the Read-Across Tool Developed

in the DTC Laboratory

The Read-Across tool, developed by the DTC Laboratory, has already been applied
for the prediction of several different activity/toxicity endpoints. The advantage of
this tool is that it provides a user-friendly GUI, which requires minimal system
specifications and storage space. Several applications in the form of case studies
have been mentioned below.

9.3.4.1 Case Study 1

Chatterjee et al. [28] utilized this tool for the quantitative predictions of nanotox-
icity in three different datasets, aiming at data gap filling. Using the Euclidean
Distance-based, Gaussian Kernel Similarity-based and Laplacian Kernel Similarity-
based similarity calculations, the authors have obtained Sensitivity, Specificity, Accu-
racy, Precision and F-measure of up to 100%, thus demonstrating the data gap-filling
ability of the tool. The hyperparameters were optimized for the computation of Gaus-
sian Kernel Similarity-based and Laplacian Kernel Similarity-based predictions in
addition to choosing the number of close source compounds along with distance and
similarity thresholds. The predicted response values thus generated were calculated
by the weighted average prediction of the individual observed response values, a
technique somewhat similar to the Consensus Model 2, as proposed by Roy et al.
9 Read-Across and RASAR Tools from the DTC Laboratory 253

[30]. The Consensus Model 1 involves the average of the predictions derived from all
the qualifying models (mean + 3 × SD) for a particular compound. The Consensus
Model 2 is derived from the weighted average predictions of all the qualifying models.
The Consensus Model 3 is obtained by selecting the model that provides the best
prediction for a particular query compound. The Read-Across-based predictions in
terms of Q 2F2 obtained in Dataset 1 were up to 0.96 using the Euclidean Distance-
based approach. For Dataset 2, the Q 2F2 values were up to 0.91 in both the Gaussian
Kernel-based and Laplacian Kernel-based predictions. The Q 2F2 values in Dataset 3
were up to 0.95, obtained by the Euclidean Distance-based approach. These values
outperformed the previous QSAR and Read-Across studies on the same datasets.

9.3.4.2 Case Study 2

Chatterjee and Roy [31] have applied both QSAR and Read-Across techniques to
predict the acute toxicities of mixtures of polar and non-polar narcotic substances
present in the environment. They have calculated 2D descriptors and adopted a 2D-
QSAR modeling technique, adhering strictly to the OECD principles, which provided
sufficient robustness, predictivity and reproducibility. For the developed PLS model,
the internal quality metrics are as follows: r 2 = 0.82 and Q 2(LOO) = 0.78, suggesting
that the model is robust, and the external validation metrics are Q 2F1 = 0.87 and
Q 2F2 = 0.87, suggesting that the model is very predictive. The authors have also
employed the Prediction Reliability Indicator tool to check the reliability of predic-
tions of a true external set. Apart from the model-derived predictions, they have also
employed a machine learning algorithm-derived similarity-based approach (Read-
Across) and utilized our Read-Across tool for toxicity predictions. They have opti-
mized the hyperparameters by dividing the training set into sub-training and sub-test
sets, and the hyperparameters which provided the best prediction of the sub-test set,
with respect to the sub-training set, were considered as the optimized hyperparam-
eters. Using these optimized settings for the hyperparameters, the predictions were
made for the original test set, with respect to the original training set. Interestingly,
in their work, the authors have reported that the external validation metrics generated
in case of Read-Across (Q 2F1 = 0.94, Q 2F2 = 0.94) were slightly higher than what
they have obtained using QSAR technique, thus showing enhanced predictivity. This
work demonstrates the importance and efficiency of similarity-based predictions over
model-derived predictions.

9.3.4.3 Case Study 3

De et al. [32] worked on the identification of molecules which can potentially act as
anti-SARS-CoV-2 drugs, using in silico approaches. They have utilized both QSAR
and Read-Across algorithms to quantitatively predict the half maximal inhibitory
concentration of the molecules. They have also used 2D descriptors and used them
254 A. Banerjee and K. Roy

to develop 2D-QSAR models. They have developed four PLS models and also
derived their consensus-based predictions. The internal quality metrics for the PLS
models were as good as r 2 = 0.672 and Q 2(LOO) = 0.612 which again suggest
sufficient robustness, and the external validation metrics in terms of Q 2F1 and Q 2F2
were reported as up to 0.839 and 0.839, respectively. The quality of predictions
was further enhanced by their consensus-based predictions, and the best predictions
in terms of the Mean Absolute Error (MAE) were obtained in case of Consensus
Model 3, where the reported values of Q 2F1 and Q 2F2 were 0.879 and 0.879, respec-
tively. The authors have also employed the similarity-based Read-Across prediction
technique. The optimization of the hyperparameters was done based on the source
compounds, and the distance and similarity-based predictions were obtained using
the optimized settings for the query chemicals. The source compounds (training set)
was first divided into sub-training and sub-test sets, and the combination of hyper-
parameters which provided the best predictions for the sub-test set, with respect
to the sub-training set, was considered as the optimized hyperparameters. This
optimized setting was then used to predict the toxicity of the original query set
compounds with respect to the original source compounds. In this work also, it has
been reported that the external validation metrics obtained in case of Read-Across-
based predictions were much better than the external validation metrics obtained
from the Partial Least Squares models, as well as their consensus-based predic-
tions. The external validation metrics for the Read-Across-based predictions were
up to Q 2F1 = 0.932 and Q 2F2 = 0.932, while the best predictions for the PLS
models were up to Q 2F1 = 0.839 and Q 2F2 = 0.839 and the consensus-based predic-
tions Q 2F1 = 0.879 and Q 2F2 = 0.879. This work also demonstrates the increased
Precision of the similarity-based predictions over model-derived predictions.

9.3.4.4 Case Study 4

Paul et al. [33] worked on the soil ecotoxicity predictions against Folsomia candida
using computational techniques. In this work, the use of Read-Across on the ecotoxi-
city predictions of Folsomia candida was performed for the first time. Two of the most
widely used in silico techniques—QSAR and Read-Across—were used to predict
the half maximal effective concentration of a set of compounds on Folsomia candida.
The authors have developed four individual PLS models which report an internal vali-
dation metrics in terms of r 2 and Q 2(LOO) as up to 0.762 and 0.633, respectively. The
reported external validation metrics were up to Q 2F1 = 0.714 and Q 2F2 = 0.642,
which suggest that the models have sufficient predictivity. The predictivity was
further enhanced by the application of a consensus-based prediction algorithm which
reported the Q 2F1 and Q 2F2 values of up to 0.726 and 0.656, respectively, based on
consensus model 3, which reported the lowest MAE value. The authors have also
performed Read-Across, an unsupervised machine learning approach, to check the
effect on the quality of predictions. They have divided the dataset into sub-training
9 Read-Across and RASAR Tools from the DTC Laboratory 255

and sub-test sets, and like the previous work, they have optimized the hyperparame-
ters based on the prediction quality of the sub-test set. The optimized setting was then
applied to the target compounds, while calculating the Read-Across-based predic-
tions, with respect to the original source compounds. It is interesting to note that in this
case also, the external validation results (Q 2F1 = 0.775 and Q 2F2 = 0.717) supersede
the ones that were obtained from the QSAR approach, even after their consensus-
based predictions. This work also proves that Read-Across-based predictions can
potentially be a more effective tool in the quantitative predictions of toxicities than
the conventional model-based QSAR approach.

9.3.4.5 Case Study 5

Banerjee et al. [34] reported in silico modeling of the androgen receptor binding
affinity of various Endocrine Disruptor Compounds (EDCs) in rats. The authors
have adopted the 2D-QSAR technique and Read-Across algorithm—two of the most
commonly used in silico approaches for the prediction of response. The 2D-QSAR
technique involved the steps like the collection of data, calculation of descriptors,
division of the dataset into training and test sets, feature selection of the essential
structural and physicochemical descriptors, development of initial MLR models and
finally the development of a PLS model. The internal quality and validation metrics
obtained in the QSAR approach were R 2 = 0.737 and Q 2(LOO) = 0.680, which
suggest that the developed model is robust. The external validation metric values
in terms of Q 2F1 and Q 2F2 were acceptable (Q 2F1 = 0.582 and Q 2F2 = 0.582). The
authors have also performed chemical Read-Across using the tool employing the
features selected in the QSAR analysis. The training set data were further divided
into sub-training and sub-test data sets, and a variety of combinations of hyperpa-
rameters were tried. The combination of the hyperparameters that provided the best
results of the sub-test set in terms of its Q 2F1 and Q 2F2 values was considered as the
optimized hyperparameters. These optimized settings of the hyperparameters were
then employed to predict the original test set compounds with respect to the original
training set compounds. The external validation metrics obtained after using the opti-
mized hyperparameters were Q 2F1 = 0.635 and Q 2F2 = 0.635. From these results,
it is evident that the predictions obtained in case of Read-Across are slightly better
than the results obtained in case of QSAR. This work again shows that chemical
Read-Across can be a potential approach for in silico predictions as an alternative to
QSAR.
The fact that the Read-Across tool can potentially predict the response values
of the query compounds, which do not have known observed response values, and
we felt that it was essential to judge the confidence measure for the predictions of
each query compound. To assess the quality of predictions of individual compounds,
Banerjee et al. [35] have adopted certain error-based and similarity-based measures
using which it is easier to identify the quality of predictions. In their work [35],
the authors have elaborated these error and similarity measures, modeled them and
identified the most important error and similarity measures using various techniques
256 A. Banerjee and K. Roy

like the mean difference among the highest and the lowest residual compounds,
Linear Discriminant Analysis of errors—a classification-based modeling technique
and Sum of Ranking Differences (SRD) [36]. The measures which were used to
assess the uncertainty in Read-Across-based predictions include SD_activity, which
defines the weighted standard deviation of the observed response values of the close
“n” source compounds to a particular query compound. The mathematical expression
of the SD_activity is given in Eq. 9.6.
[
|∑
| n w (x − x ) 2
| i=1∑i i wtd n
sweighted = n × , (9.6)
i=1 wi n−1
∑n
wi xi
xwtd = ∑i=1n , (9.7)
i=1 wi
(∑n )2
i=1 wi
n = ∑n ( 2 ) . (9.8)
i=1 wi

The expression wi denotes the similarity weightage, xwtd signifies the weighted
average prediction and n stands for the effective degree of freedom. CV_activity
defines the coefficient of variation of the observed response values. It can be denoted
mathematically as Eq. 9.9:
sweighted
CVactivity = . (9.9)
xwtd

The Euclidean Distance-based similarity function can be described as the simi-

larity value between two compounds that is obtained from their Euclidean Distance.
It is expressed as in Eq. 9.10.

f (E D) = 1 − d(X, Y ). (9.10)

In this equation, d(X, Y ) denotes the Euclidean Distance between two compounds
and f (E D) signifies the Euclidean Distance-based similarity. Another similarity-
based measure is the Gaussian Kernel-based Similarity function, which utilizes the
L 2 norm of the Euclidean Distance, i.e., squared Euclidean Distance. The Gaussian
Kernel Similarity and Laplacian Kernel Similarity were previously defined in Eqs.
(9.2) and (9.3).
The term average similarity denotes the mean value of the similarities of all the
close “n” source compounds, selected for each query compound. It can demonstrate
the closeness of the source compounds to the target/query compound. The expression
for the computation of average similarity is demonstrated in Eq. 9.11.
∑n
fi
Similarityaverage = i=1
. (9.11)
n
9 Read-Across and RASAR Tools from the DTC Laboratory 257

The term f i in this expression is the individual similarity values of the close
“n” source compounds, with respect to the target compound. A dispersion measure,
which is essential to estimate the uncertainty in Read-Across predictions, is the
Standard Deviation of the Similarity values (SD_similarity) of the close “n” source
compounds, with respect to a particular query compound. This is essential to
check the dispersion of the similarity values among the selected close “n” source
compounds. Mathematically, SD_similarity can be denoted as Eq. 9.12.
/
∑n 2
i=1 (f − f)
ssimilarity = . (9.12)
n−1

In Eq. 9.12, Ssimilarity denotes the SD_similarity, while f is the average similarity
of the close “n” source compounds. Another similarity measure is MaxPos, which
signifies the similarity value of the closest source compound, with respect to the
target compound, and having an observed response value greater than the threshold
(mean observed response values of all the source compounds). This measure is essen-
tial to estimate the closeness of a particular query compound toward positive or
negative (with respect to the threshold observed response). Likewise, MaxNeg signi-
fies the similarity value of the closest source compound, with respect to the target
compound, and having an observed response value lower than the threshold. Again,
MaxNeg provides the information on how close the query compound is to the nega-
tive source congeners. The metric Abs(MaxPos-MaxNeg) is the absolute differences
in the MaxPos and MaxNeg similarity values, and a high value indicates that a
particular query compound is significantly more similar to the positive or negative
source compounds. A concordance measure g has been applied [25], which takes
into account the fraction of close source compounds that have a higher observed
response value than the threshold (Positive Fraction). The mathematical equation for
the calculation of g has already been discussed in Eq. 9.4, and its value ranges from
0 to 1. The summary of these similarity and error measures has been provided in
Table 9.1.
From the analysis of all the similarity and error-based measures, the authors have
set the following criteria for the estimation of the uncertainty in Read-Across-based
predictions in Table 9.2.
Reliability estimates: Very Good (All criteria met); Good (Criterion 1 and at least
one of the rest, but not all); Moderate (Any one met); Bad (None of the criteria met).

9.4 Read-Across Structure–Activity Relationship—A Novel

Concept

So far, in silico approaches for the assessment of activity/property/toxicity have

centered around the use of QSAR and various Machine Learning (ML) approaches
like Support Vector Machine (SVM) and Artificial Neural Networks (ANN).
258 A. Banerjee and K. Roy

Table 9.1 List of similarity and various error measures generated for each query compound during
Read-Across predictions
Measures Description Comment
SD_activity (sweighted ) Weighted standard deviation of the observed Dispersion
response values of the close “n” source measure
compounds for each query compound
CV_activity Coefficient of variation of the response Relative error
measure
Euclidean distance-based It determines the similarity between two Similarity
similarity function compounds X and Y using the Euclidean function ( f )
distance approach
Gaussian Kernel-based It determines the similarity between two Similarity
similarity function compounds X and Y using the Gaussian Kernel function ( f )
similarity approach
Laplacian Kernel-based It determines the similarity between two Similarity
similarity function compounds X and Y using the Laplacian function ( f )
Kernel Similarity approach
Average similarity Mean similarity to the selected close source Similarity
compounds for each query compound measure
SD_similarity Standard deviation of the similarity values of Dispersion
the selected close source compounds for each measure
query compound
MaxPos Maximum similarity level to the positive close Similarity
source set compounds (based on the “training measure
set” observed mean)
MaxNeg Maximum similarity level to the negative close Similarity
source set compounds (based on the “training measure
set” observed mean)
AbsDiff or Absolute difference between MaxPos and Similarity
Abs(MaxPos-MaxNeg) MaxNeg measure
g [25] This is a concordance measure Similarity
measure

Table 9.2 Estimation of the

Criteria Dispersion/Similarity measures Desired range
reliability of
Read-Across-based 1 SD_activity (Euclidean) ≤0.75
predictions by the levels of 2 g (Euclidean) ≤0.4a
the similarity/dispersion
measures [35] 3(a) Average similarity (Euclidean) ≥0.85
3(b) CV_similarity (Euclidean) ≥0.05
a Corresponds to PosFrac ≥0.8 or PosFrac ≤0.2
9 Read-Across and RASAR Tools from the DTC Laboratory 259

Recently, there has been a rise in the adoption of similarity-based approaches like
Read-Across mainly for their simplicity and Accuracy of the predictions, as esti-
mated from various external validation metrics. Read-Across has now become one
of the most useful algorithms for data gap filling, especially in cases where there
is a scarcity of experimental data, as this technique does not involve the develop-
ment of a model, and thus, accurate predictions can be obtained based on similarity
values. Luechtefeld et al. [24] in 2018 proposed the idea of Read-Across Structure–
Activity Relationship which combines the Read-Across algorithm with the QSAR
methodology. They adopted a machine learning approach and used the concept of
MaxPos and MaxNeg (vide supra) to develop classification-based models. In a recent
work [29], Banerjee and Roy have tried to club the advantages of QSAR modeling
and Read-Across approach and derived a novel quantitative Read-Across Structure–
Activity Relationship (q-RASAR) approach. The authors have also tried a variety
of similarity and error-based descriptors in the generation of RASAR models unlike
Luechtefeld et al., where they have only used the maximum similarity values with
the positive and negative source compounds.

9.5 The RASAR Descriptor Calculator Tool from the DTC

Laboratory

The RASAR Descriptor Calculator tool, developed by the DTC Laboratory, is a

simple Java-based software application, which quickly computes the similarity and
error-based measures for each particular compound that can be used for the develop-
ment of RASAR models. We recommend to use this software only after successful
Read-Across-based predictions with the optimized hyperparameters. These opti-
mized hyperparameters are also taken as inputs in the RASAR descriptor calculator
tool, based on which the RASAR descriptors are computed. During its execution, the
tool asks the user for the similarity-based methods based on which it will compute
the descriptors. The similarity-based measures which the user can select are the
Euclidean Distance-based, the Gaussian Kernel-based and the Laplacian Kernel-
based measures. If the user selects the Euclidean Distance-based approach, the tool
asks for the number of similar training compounds the tool will consider and the
threshold value of the distance. If the user selects the Gaussian Kernel Similarity-
based approach, the tool asks for the σ value (an optimized hyperparameter), the
number of similar training compounds and the threshold value for similarity. Selec-
tion of the Laplacian Kernel Similarity-based approach prompts the tool to ask for
the γ value (another optimizable hyperparameter), the number of similar training
compounds and the threshold value for similarity. In each case, the program gener-
ates two output files, namely TESTsetfilename_Sort.xlsx and TESTsetfilename_
RASAR_Descriptors.xlsx. The sort file contains the sorted values of the similarities
for each query compound, with all the source compounds, according to the similarity
measure specified by the user. The RASAR descriptor file contains the descriptors for
260 A. Banerjee and K. Roy

the development of RASAR models. Apart from the previously discussed similarity
and error-based uncertainty measures, this tool also computes five new measures.
The first of these new measures is the product of the Banerjee-Roy coefficient and
the average similarity of the close “n” source compounds (gm × Avg.Sim), and the
second is the product of the Banerjee-Roy coefficient and the Standard Deviation
of the similarity values of the close “n” source compounds (gm × SD_similarity).
The other two measures include the average similarity of the compounds, consti-
tuting the close “n” neighbors having a response value greater than the threshold
(Pos.Avg.Sim), and the average similarity of the close “n” source compounds having
a response value lower than the threshold (Neg.Avg.Sim). Lastly, the RA_function
is derived from Read-Across, which acts like a composite variable and contains
all the information of the structural and physicochemical descriptors selected to
perform Read-Across initially. The tool is freely available and can be downloaded
from https://sites.google.com/jadavpuruniversity.in/dtc-lab-software/home.

9.5.1 Pre-Requisites for Using This Tool

9.5.1.1 System Specifications

Like the Read-Across tool, RASAR-Desc-Calc-v1.0 (Note: the current version is

RASAR-Desc-Calc-v3.0.1) does not require exhaustive system resources, i.e., it can
run on computers having standard memories for RAM and HDD/SSD. However,
since it is a Java-based software tool, it is necessary that a particular system needs
to have Java installed before running this tool. The link for downloading JDK has
already been mentioned above.

9.5.1.2 Input File Specifications

The input file specifications for the RASAR Descriptor Calculator tool is the same
as the Read-Across tool. The training and test set files must bear the extension of
.xlsx, and in each of these files, the compound number constitutes the first column,
the descriptors in subsequent columns and the observed response values at the last
column.

9.5.2 Downloading and Execution of the Tool

• The software in the form of a .zip file has been made available at the DTC lab tools
supplementary webpage (https://sites.google.com/jadavpuruniversity.in/dtc-lab-
software/home)
9 Read-Across and RASAR Tools from the DTC Laboratory 261

• The user will need to download the.zip file and extract the contents. The folder
consists of an executable .jar file, a library folder and sample training and test set
files in.xlsx format.
• The training and test set files need to be placed inside this RASAR-Desc-Calc-v1.0
folder, i.e., the same folder which contains the.jar file.
• The user needs to double click on the executable.jar file and the program will be
executed.
• The user needs to enter 1, 2, or 3 based on the similarity measure using which the
user wants the descriptors to be calculated. If the user enters 1, the calculations
will be based on the Euclidean Distance-based approach and the program will only
take input of the number of close source compounds and the distance threshold. If
the user enters 2, the calculations will be based on the Gaussian Kernel Similarity-
based approach, which asks the user to enter the value for σ, the number of close
source compounds and the similarity threshold. If the user enters 3, the calculation
will be based on Laplacian Kernel-based Similarity, where the system asks for the
value of γ , the number of close source compounds and the similarity threshold.
• The sorted similarity measures will be printed to TESTsetfilename_
sort.xlsx, while the descriptors will be printed to TESTsetfilename_RASAR_
Descriptors.xlsx.
• Using these descriptors, one can go for RASAR model development with or
without involvement of the original structural and physicochemical descriptors.

9.5.3 Analysis of the Output Files

The above tool generates two different output files, namely TESTsetfilename_
sort.xlsx, which prints the sorted similarity values of the source compounds according
to the selected similarity-based measure, along with their observed response values,
and TESTsetfilename_RASAR_Descriptors.xlsx, which prints the computed simi-
larity and error measures which are used as RASAR descriptors for the generation
of RASAR models.

9.5.4 Application of the RASAR Descriptor Calculator Tool

Developed by the DTC Laboratory

Banerjee and Roy [29] have recently demonstrated the application of the q-RASAR
approach taking a case study of androgen receptor binding affinity. With the
novel idea of utilizing the similarity and error measures obtained from Read-
Across-v4.1 (https://sites.google.com/jadavpuruniversity.in/dtc-lab-software/home)
as descriptor values, the authors have clubbed the structural and physicochemical
descriptors (obtained after feature selection) with the similarity and error measures.
The total set of descriptors thus obtained was further subjected to feature selection,
262 A. Banerjee and K. Roy

and various MLR models were generated based on the combination of these descrip-
tors. Noise and intercorrelation among the descriptors were removed by generating
Partial Least Squares models, and then, the prediction quality was further enhanced
by consensus-based predictions.
The androgen receptor binding affinity data of various molecules on rats
were collected from the Endocrine Disruptor Knowledge Base (EDKB) database
(https://www.fda.gov/science-research/endocrine-disruptor-knowledge-base/access
ing-ar-binding-dataset-androgen-receptor), and the descriptors were generated. The
intercorrelated descriptors were removed by the technique of data pre-treatment.
After removal of the intercorrelated descriptors, due to the absence of a true
external set, the dataset was divided into training and test sets based on a certain
pre-defined algorithm (Euclidean Distance-based division). Thereafter, feature
selection algorithms like Genetic Algorithm-MLR and Best Subset Selection were
employed, and consequently, the authors obtained a certain set of structural and
physicochemical descriptors which were believed to contribute significantly toward
the prediction of the androgen receptor binding affinity. Using these selected
descriptors, the training set was further divided into sub-training and sub-test
sets, and a series of Read-Across-based predictions were obtained, changing the
hyperparameters in each case. The set of hyperparameters, which provided the best
predictions for the sub-test set, in terms of the external validation metrics and the
MAE, were used to perform Read-Across-based predictions for the original test
set (query set), using the original training set compounds (source compounds).
Although the initial concept of Read-Across demonstrates an unsupervised learning
approach, the process of optimization of the hyperparameters in this work can be
considered a supervised learning approach. The final output file generated from
Read-Across-v4.1 had the predicted response values, the external validation metrics,
the overall error measures (RMSEP and MAE) and the similarity and error measures
for each query compound. The authors have utilized some of these similarity and
error measures, namely SD_activity, CV_activity, average similarity, SD_similarity,
MaxPos, MaxNeg and Abs(MaxPos-MaxNeg) (Table 9.1), and chosen them as the
RASAR descriptors. These RASAR descriptors were then clubbed to the previously
selected structural and physicochemical descriptors to obtain the descriptor pool.
The complete descriptor pool was then subjected to feature selection using the Best
Subset Selection technique—an approach which generates multiple MLR models
from all possible combinations of descriptors. The best four MLR models were
chosen based on their internal and external validation metrics, and four individual
PLS models were developed. Also, three pooled PLS models were developed
which consisted of the pooled descriptor combinations from the individual PLS
models. To enhance the prediction quality, the authors have utilized an intelligent
consensus-based prediction technique [30]. The consensus model 3, which utilizes
the best prediction of a particular compound from a selected model out of all the
available PLS models, showed the best prediction quality. Figure 9.6 demonstrates
the workflow followed by Banerjee and Roy in their work.
The authors have then observed that the concordance measure g, proposed by Wu
et al. [25], is not able to distinguish between the positive and the negative query
9 Read-Across and RASAR Tools from the DTC Laboratory 263

Fig. 9.6 Workflow of the q-RASAR methodology [29]

compounds, since it only takes the PosFrac into account (however, please note g
was used in the original work in a different context). Therefore, according to the
formula of g, a compound having PosFrac of 0.6 has the same value of g with
another compound having a PosFrac of 0.4. Likewise, identical values of g can also
be obtained in cases where PosFrac is 0.3 and 0.7, 0.2 and 0.8, 0.1 and 0.9, 0 and
1. Also, a compound may have a higher PosFrac value, but may possess a higher
level of similarity to the negative source compounds and vice versa. To address all
these aspects, the authors have developed novel Banerjee-Roy coefficient (gm ) which
utilizes both the PosFrac and the MaxPos/MaxNeg values. Equation 9.5 denotes the
mathematical form of this novel coefficient. The value of n in the equation is equal to
1, when MaxNeg > MaxPos, while n = 2 when MaxNeg < MaxPos. This incorporates
a directionality among the query compounds as the compound which is more likely
to belong to the negative class has a negative value of g, while a compound which
has a tendency to be positive has a positive value of g. Using this modified g (gm ), the
authors have re-developed one of the pooled PLS models, and it was observed that
both the internal and external validation metrics obtained in this model were better
than all the previous QSAR models and Read-Across (only external validation). The
Mean Absolute Error obtained was even lower than what the authors have obtained
in the previous consensus-based model. The summary of the internal and external
validation metrics has been tabulated in Table 9.3.
 264

Table 9.3 Summary of the internal and external validation metrics obtained in our work and their comparison with previous works (the best values of different
metrics are shown in bold) [29]
PLS model(s) LVs R2 Q 2(LOO) Q 2F1 Q 2F2 Q 2F3 MAE -Fitted(Train) MAE -LOO(Train) M AE (T est)
Pooled descriptor models
P1 (M1 + M2) 3 0.718 0.666 0.671 0.670 0.683 0.479 0.517 0.478
P2 (M1 + M2 + M3) 2 0.754 0.718 0.630 0.629 0.644 0.441 0.470 0.504
P1m (Using gm ) 4 0.753 0.698 0.674 0.674 0.686 0.434 0.477 0.461
Intelligent consensus model
ICP3 (M1 + M2 + M3 + – – – 0.652 0.652 0.665 – – 0.463
M4) (CM3)
Previous 2D-QSAR model and Read-Across predictions (Banerjee et al. [34])
2D-QSAR 3 0.737 0.680 0.582 0.582 0.606 0.456 0.497 0.539
Quantitative read-across – – – 0.635 0.635 0.656 – – 0.468
(Gaussian Kernel
Similarity-based)
Previous works done by other researchers
3D-QSAR (CoMFA) by – 0.902 0.571 – – – – – –
Hong et al. [37]
nTraining = 146 nTest = 8
Classification-based QSAR – – – – – – – – –
by Piir et al. [38]
nTraining = 1688 nTest =
5273
A. Banerjee and K. Roy
9 Read-Across and RASAR Tools from the DTC Laboratory 265

9.6 Conclusion

In compliance with the regulatory authorities like EU-REACH, in silico approaches

of activity/toxicity estimation have had a surge in its applications in various fields.
The most important aspect of this approach is that it avoids animal testing, experi-
mental and instrumental errors and also requires minimal time to obtain accurate and
reliable results. 2D-QSAR involves simple, transferable and interpretable models,
while higher dimensional QSAR approaches deal with the spatial arrangement of
atoms and molecules and involve various other steps like conformational analysis and
alignment, which require exhaustive system resources, and reproducibility is compro-
mised. The recent trend is to follow similarity-based prediction techniques that do
not require a model to predict the response values of the compounds constituting
the external set. In cases, where the number of training data points is very limited,
model-derived predictions are more likely to produce biased or erroneous results due
to an insufficient degree of freedom, but the similarity-based approaches like Read-
Across can still be able to generate reliable predictions. Thus, the similarity-based
approaches like Read-Across are very useful in data gap filling. In our Read-
Across tool (https://sites.google.com/jadavpuruniversity.in/dtc-lab-software/home),
we provide the predictions based on the Euclidean Distance approach, the Gaussian
Kernel Similarity approach and the Laplacian Kernel Similarity approach and also
the external validation metrics in terms of Q 2F1 and Q 2F2 along with the overall error
measures in terms of RMSEP and MAE. This tool also generates certain measures
using which one can estimate the uncertainty in the Read-Across predictions if the
observed response values of the query compounds are not available. Our Read-
Across approach deals with the local similarities with the close source compounds
for each of the query compound, and thus, one may expect a slightly better predic-
tion for the query compounds. The concept of RASAR, which was already intro-
duced by Luechtefeld et al. [24], brings together the concept of Read-Across and
QSAR. They have adopted machine learning approaches to develop classification-
based RASAR models. We at DTC Laboratory performed quantitative Read-Across
Structure–Activity Relationship (q-RASAR) and generated quantitative predictions
by combining the advantages of Read-Across and QSAR. It was observed that the
predictions obtained from q-RASAR were better in terms of both the internal and
external validation metrics than majority of the work that have been done previ-
ously on the androgen receptor binding affinity of endocrine disruptors. Thus, we
have developed a novel RASAR descriptor calculator tool (https://sites.google.com/
jadavpuruniversity.in/dtc-lab-software/home) for the quick and efficient calculation
of similarity and error-based descriptors to develop q-RASAR models. The devel-
opment of data fusion RASAR models and linking them with multiple Molecular
Initiating Events (MIEs) and Adverse Outcome Pathways (AOPs) can potentially be
a useful algorithm for future drug discovery and development. We also believe that
the Read-Across and q-RASAR models have a lot of potential for bridging data gaps,
and probably, they may prove to be the essential prediction tools for the future.
266 A. Banerjee and K. Roy

Acknowledgements AB thanks Jadavpur University, Kolkata, for a scholarship. KR thanks the

Science and Engineering Research Board (SERB), New Delhi, for financial assistance under the
MATRICS scheme (MTR/2019/000008).

References

1. Mech A, Rasmussen K, Jantunen P, Aicher L, Alessandrelli M, Bernauer U, Bleeker EAJ, Bouil-

lard J, Fanghella PDP, Draisci R, Dusinska M, Encheva G, Flament G, Haase A, Handzhiyski
Y, Herzberg F, Huwyler J, Jacobsen NR, Jeliazkov V, Jeliazkova N, Nymark P, Grafström R,
Oomen AG, Polci ML, Sandström CRJ, Shivachev B, Stateva S, Tanasescu S, Tsekovska R,
Wallin H, Wilks MF, Zellmer S, Apostolova MD (2019) Insights into possibilities for grouping
and read-across for nanomaterials in EU chemicals legislation. Nanotoxicology 13(1):119–141.
https://doi.org/10.1080/17435390.2018.1513092
2. Fischer I, Milton C, Wallace H (2020) Toxicity testing is evolving! Toxicol Res 9(2):67–80.
https://doi.org/10.1093/toxres/tfaa011
3. Hemmerich J, Ecker FG (2020) In silico toxicology: From structure–activity relationships
towards deep learning and adverse outcome pathways. WIRES Comp Mol Sci 10:e1475.
https://doi.org/10.1002/wcms.1475
4. Gomes SIL Scott-Fordsmand JJ Amorim MJB (2021) Alternative test methods for (nano)
materials hazards assessment: Challenges and recommendations for regulatory preparedness.
Nano Today 40:101242. https://doi.org/10.1016/j.nantod.2021.101242
5. Nymark P, Bakker M, Dekkers S, Franken R, Fransman W, García-Bilbao A, Greco D, Gulu-
mian M, Hadrup N, Halappanavar S, Hongisto V, Hougaard KS, Jensen KA, Kohonen P,
Koivisto AJ, Maso MD, Oosterwijk T, Poikkimäki M, Rodriguez-Llopis I, Stierum R, Sørli JB,
Grafström R (2020) Toward rigorous materials production: new approach methodologies have
extensive potential to improve current safety assessment practices. Small 16:1904749. https://
doi.org/10.1002/smll.201904749
6. Madden JC, Enoch SJ, Paini A, Cronin MTD (2020) A review of in silico tools as alternatives
to animal testing: principles, resources and applications. Alt Lab Ani 48(4):146–172. https://
doi.org/10.1177/0261192920965977
7. Gellatly N, Sewell F (2019) Regulatory acceptance of in silico approaches for the safety assess-
ment of cosmetic-related substances. Comp Toxicol 11:82–89. https://doi.org/10.1016/j.com
tox.2019.03.003
8. Mangiatordi GF, Alberga D, Altomare CD, Carotti A, Catto M, Cellamare S, Gadaleta D,
Lattanzi G, Leonetti F, Pisani L, Stefanachi A, Trisciuzzi D, Nicolotti O (2016) Mind the gap!
a journey towards computational toxicology. Mol Inf 35:294–308. https://doi.org/10.1002/
minf.201501017
9. Kovarich S, Ceriani L, Gatnik MF, Bassan A, Pavan M (2019) Filling data gaps by read-across:
a mini review on its application, developments and challenges. Mol Inf 38:1800121. https://
doi.org/10.1002/minf.201800121
10. Hartung T (2016) Making big sense from big data in toxicology by read-across. ALTEX 33(2).
https://doi.org/10.14573/altex.1603091
11. Maldonado AG, Doucet JP, Petitjean M, Fan B (2006) Molecular similarity and diversity in
chemoinformatics: from theory to applications. Mol Divers 10:39–79. https://doi.org/10.1007/
s11030-006-8697-1
12. Ball N, Madden J, Paini A, Mathea M, Palmer AD, Sperber S, Hartung T, van Ravenzwaay B
(2020) Key read across framework components and biology based improvements. Mutat Res
Genet Toxicol Environ 853:503172. https://doi.org/10.1016/j.mrgentox.2020.503172
13. Benfenati E, Chaudhry Q, Gini G, Dorne JL (2019) Integrating in silico models and read-across
methods for predicting toxicity of chemicals: a step-wise strategy. Environ Int 131:105060.
https://doi.org/10.1016/j.envint.2019.105060
9 Read-Across and RASAR Tools from the DTC Laboratory 267

14. Ellison CM, Enoch SJ, Cronin MTD (2011) A review of the use of in silico methods to predict
the chemistry of molecular initiating events related to drug toxicity. Expert Opin Drug Metabol
Toxicol 7(12):1481–1495. https://doi.org/10.1517/17425255.2011.629186
15. Enoch SJ, Cronin MTD, Schultz TW, Madden JC (2008) Quantitative and mechanistic read
across for predicting the skin sensitization potential of alkenes acting via michael addition.
Chem Res Toxicol 21:513–520. https://doi.org/10.1021/tx700322g
16. Schuurmann G, Ebert RU, Kuhne R (2011) Quantitative read-across for predicting the acute
fish toxicity of organic compounds. Environ Sci Technol 45:4616–4622. https://doi.org/10.
1021/es200361r
17. Kühne R, Ebert RU, von der Ohe PC, Ulrich N, Brack W, Schüürmann G (2013) Read-Across
prediction of the acute toxicity of organic compounds toward the water flea daphnia magna.
Mol Inf 32:108–120. https://doi.org/10.1002/minf.201200085
18. Russo DP, Strickland J, Karmaus AL, Wang W, Shende S, Hartung T, Aleksunes LM, Zhu H
(2019) Nonanimal models for acute toxicity evaluations: applying data-driven profiling and
read-across. Environ Health Pers 127(4):047001. https://doi.org/10.1289/EHP3614
19. Low Y, Sedykh A, Fourches D, Golbraikh A, Whelan M, Rusyn I, Tropsha A (2013) Integrative
chemical-biological read-across approach for chemical hazard classification. Chem Res Toxicol
26:1199–1208. https://doi.org/10.1021/tx400110f
20. van Ravenzwaay B, Sperber S, Lemke O, Fabian E, Faulhammer F, Kamp H, Mellert W,
Strauss V, Strigun A, Peter E, Spitzer M, Walk T (2016) Metabolomics as read-across tool: a
case study with phenoxy herbicides. Regulat Toxicol Pharmacol 81:288–304. https://doi.org/
10.1016/j.yrtph.2016.09.013
21. Przybylak KR, Schultz TW, Richarz AN, Mellor CL, Escher SE, Cronin MTD (2017) Read-
across of 90-day rat oral repeated-dose toxicity: a case study for selected β-olefinic alcohols.
Comp Toxicol 1:22–32. https://doi.org/10.1016/j.comtox.2016.11.001
22. Schultz TW, Richarz AN, Cronin MTD (2019) Assessing uncertainty in read-across: questions
to evaluate toxicity predictions based on knowledge gained from case studies. Comp Toxicol
9:1–11. https://doi.org/10.1016/j.comtox.2018.10.003
23. Alves VM, Golbraikh A, Capuzzi SJ, Liu K, Lam WI, Korn DR, Pozefsky D, Andrade CH,
Muratov EN, Tropsha A (2018) Multi-descriptor read across (MuDRA): a simple and trans-
parent approach for developing accurate quantitative structure–activity relationship models. J
Chem Inf Model 58(6):1214–1223. https://doi.org/10.1021/acs.jcim.8b00124
24. Luechtefeld T, Marsh D, Rowlands C, Hartung T (2018) Machine learning of toxicological big
data enables read-across structure activity relationships (RASAR) outperforming animal test
reproducibility. Toxicol Sci 165(1):198–212. https://doi.org/10.1093/toxsci/kfy152
25. Wu J, D’Ambrosi S, Ammann L, Stadnicka-Michalak J, Schirmer K, Baity-Jesi M (2022)
Predicting chemical hazard across taxa through machine learning. Environ Int 163:107184.
https://doi.org/10.1016/j.envint.2022.107184
26. AbdulHameed MDM, Liu R, Schyman P, Sachs D, Xu Z, Desai V, Wallqvist A (2021) ToxPro-
filer: toxicity-target profiler based on chemical similarity. Comp Toxicol 18:100162. https://
doi.org/10.1016/j.comtox.2021.100162
27. Manganelli S, Benfenati E (2016) Use of read-across tools. In: Benfenati E (ed) In silico
methods for predicting drug toxicity. Humana Press, pp 305–322. https://doi.org/10.1007/978-
1-4939-3609-0_13
28. Chatterjee M, Banerjee A, De P, Gajewicz A, Roy K (2022) A novel quantitative read-across tool
designed purposefully to fill the existing gaps in nanosafety data. Env Sci: Nano. 9:189–203.
https://doi.org/10.1039/D1EN00725D
29. Banerjee A, Roy K (2022) First report of q-RASAR modeling towards an approach of easy
interpretability and efficient transferability. Mol Divers 26(5):2847–2862. https://doi.org/10.
1007/s11030-022-10478-6
30. Roy K, Ambure P, Kar S, Ojha PK (2018) Is it possible to improve the quality of predictions
from an “intelligent” use of multiple QSAR/QSPR/QSTR models? J Chemom 32:e2992.
https://doi.org/10.1002/cem.2992
268 A. Banerjee and K. Roy

31. Chatterjee M, Roy K (2022) Application of cross-validation strategies to avoid overestimation

of performance of 2D-QSAR models for the prediction of aquatic toxicity of chemical mixtures.
SAR QSAR Env Res 33(6):463–484. https://doi.org/10.1080/1062936X.2022.2081255
32. De P, Kumar V, Kar S, Roy K, Leszczynski J (2022) Repurposing FDA approved drugs as
possible anti-SARS-CoV-2 medications using ligand-based computational approaches: sum of
ranking difference-based model selection. Struc Chem. https://doi.org/10.1007/s11224-022-
01975-3
33. Paul R, Chatterjee M, Roy K (2022) First report on soil ecotoxicity prediction against Folsomia
candida using intelligent consensus predictions and chemical read-across. Env Sci Pollut Res.
https://doi.org/10.1007/s11356-022-21937-w
34. Banerjee A, De P, Kumar V, Kar S, Roy K (2022) Quick and efficient quantitative predictions
of androgen receptor binding affinity for screening endocrine disruptor chemicals using 2D-
QSAR and chemical read-across. Chemosphere 309:136579. https://doi.org/10.1016/j.chemos
phere.2022.136579
35. Banerjee A, Chatterjee M, De P, Roy K (2022) Quantitative predictions from chemical read-
across and their confidence measures. Chemom Intell Lab Syst 227:104613. https://doi.org/10.
1016/j.chemolab.2022.104613
36. Heberger K (2010) Sum of ranking differences compares methods or models fairly. TrAC
Trends Anal Chem 29(1):101–109. https://doi.org/10.1016/j.trac.2009.09.009
37. Hong H, Fang H, Xie Q, Perkins R, Sheehan DM, Tong W (2003) Comparative molecular
field analysis (CoMFA) model using a large diverse set of natural, synthetic and environmental
chemicals for binding to the androgen receptor. SAR QSAR Env Res 14(5–6):373–388. https:/
/doi.org/10.1080/10629360310001623962
38. Piir G, Sild S, Maran U (2021) Binary and multi-class classification for androgen receptor
agonists, antagonists and binders. Chemosphere 262:128313. https://doi.org/10.1016/j.chemos
phere.2020.128313
Chapter 10
Databases for Drug Discovery
and Development

Supratik Kar and Jerzy Leszczynski

Abstract Computational drug design and discovery have taken center stage atten-
tion during the time of COVID-19. The science community acknowledges the impor-
tance of ligand-based drug design (LBDD) and structure-based drug design (SBDD)
to nullify the problem associated with a typical drug discovery process. In the modern
era, a complement between experimental, theoretical, and computational approaches
can make the drug discovery process rational, economical, and fast. Undoubtedly,
computational power has increased manifold compared to the last few decades,
making it possible to run many unthinkable calculations that cannot be imagined
a few years ago. Along with the computational power, resources like open-access
and commercial organic chemicals, phytochemicals, approved, experimental and
investigational drugs, peptides, and metabolomic databases have increased enor-
mously. Compared to designing a new drug, utilizing existing chemical and drug
databases for virtual screening makes the process faster as the database chemicals
are already synthesized (in most cases) and characterized. Even in a few instances,
absorption, distribution, metabolism, excretion, and toxicity (ADMET) profiles are
checked along with data for preclinical and clinical trials (primarily for investiga-
tional and/or in the process of approval drugs). A drug database is also a powerful
resource for drug repurposing, where an old, approved drug for a specific disease
can be used to treat another common/new/rare disease. The idea is increasingly
becoming an attractive proposition as it comprises the use of already evaluated de-
risked compounds which help lower the new drug development costs in a shorter
time. Therefore, drug databases have an immense role to play as a repository of
potential drugs for any common to a rare disease in the process of CADD and for
the experimental scientists.

S. Kar (B)
Department of Chemistry, Chemometrics and Molecular Modeling Laboratory, Kean University,
1000 Morris Avenue, Union, NJ 07083, USA
e-mail: [email protected]
J. Leszczynski
Department of Chemistry, Physics and Atmospheric Sciences, Interdisciplinary Center
for Nanotoxicity, Jackson State University, Jackson, MS 39217, USA

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 269
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7_10
270 S. Kar and J. Leszczynski

Keywords Database · Drug design · Drug discovery · Biological activity · Targets

10.1 Introduction

Computer-aided drug designing (CADD) combines multiple computational

approaches, decreasing the drug discovery process timeline manifold. Computational
and literature resources are the backbone of the CADD [1]. With the advancement
of computational resources, one can perform unimaginable calculations which could
not be considered a few decades ago. Parallelly, storage capacity of webservers and
cloud storage helped us to have a single platform of databases where researchers can
have access to millions to billions of chemicals irrespective of organic chemicals,
approved, investigational and experimental drugs, phytochemicals, peptides as well
as databases for metabolomics and therapeutics targets too [2–4].
Standard drug design to discovery process [5] is a multistep process which is
illustrated in Fig. 10.1. The most distinct steps are shown here where in each step, a
different form of database can be utilized which shows the importance of databases
in drug discovery [6, 7]. Based on the discovery step, researcher has to select the
best possible database or combination of multiple databases for the study.

Fig. 10.1 Role of different databases in CADD

10 Databases for Drug Discovery and Development 271

The databases are always not only limited to small organic molecules, but they
can also be sometime peptides and biologicals molecules too. Databases are the
key resources of any form off bioinformatics and cheminformatics project. These
databases can be extremely big regarding the number of available molecules and
result from text mining and automatic processes, whereas others can contain highly
curated data. In recent time, a huge number of fairly large databases are freely
available to use which has open more opportunists for open-access drug design and
discovery process. The databases can be used entirely by downloading them or in
many cases there is an option to search specific class and/or structural scaffolds from
the big chunk of molecules. Therefore, depending on the requirements, a researcher
can use the database for the study. A molecular cloud is represented in Fig. 10.2
using a small section of molecules (of SuperNatural II database only 150 from pool
of 325,508 compounds) [8] to show the variety of structural scaffolds exist in a
database.
Many times, these databases are directly connected with modeling and diverse
analysis tools like docking, toxicity prediction, BLAST, etc. which help to utilize
initial assessment of the database for the specific study. An ideal database should
contain chemical structure information in form of SMILES/SDF, major physico-
chemical properties, any form of experimental data related to activity and/or toxicity

Fig. 10.2 Representation of a small section of chemicals from SuperNatural II database in form of
molecular cloud
272 S. Kar and J. Leszczynski

Fig. 10.3 Use of drug databases in published peer-reviewed research publications over the last
120 years according to Scopus

and potential vendors [9, 10]. If the molecules present in the database are already
synthesized and characterized, it is good to have them in the study as if the molecule
evolved as a potential drug, for experimental assay researcher can simply buy it
from vendor for the further study which does not require any form of synthesis or
initial characterization. A simple search of ‘drug’ and ‘database’ in Scopus can show
immense growth of drug databases and their application in drug discovery where one
can see sharp jump of the usage of databases from the year 2000 and it is reaching
all time high at present time (Fig. 10.3).

10.2 Types of Databases for Drug Discovery

Drug databases can be classified according to the chemical nature of the drug
molecule, disease-specific, target-oriented, and metabolomic pathways. Again, small
chemical molecules can be classified as investigational and experimental drugs, while
approved drugs are generally classified under drug molecules which are majorly used
for drug repurposing. This chapter discusses databases for small chemical molecules,
drug molecules, metabolomic, peptides, and therapeutic target information, as illus-
trated in Fig. 10.4. The researcher has to decide the types and requirement of the
databases for the research. If researcher wants to do virtual screening, then going
for multiple databases is a good option. Now, what types of databases one has to
use thats completely depend on the requirement of the researcher [11–13]. On the
other side, it is always better to take US FDA-approved drugs, investigational and
experimental drugs under the radar of USFDA or any govt. approved agencies for
drug repurposing purposes. Thus, selection of database is always important as this
is the first step of CADD.
10 Databases for Drug Discovery and Development 273

Fig. 10.4 Types of

databases for drug discovery

10.3 Databases

10.3.1 Chemical Molecules Database

Chemical molecules databases are the essential and most powerful resources for
virtual screening study. These databases can be used for ligand- and structure-
based drug design and discovery, where quantitative structure–activity relationships
(QSARs) [14] and machine learning models [15] can be strategically used for ligand-
based drug discovery employing these databases. On the contrary, docking, pharma-
cophore, molecular dynamics [16–18] followed by ADMET profiling studies [19,
20] can be strategically used for virtual screening of chemicals databases. For better
understanding, small chemical molecules databases are classified under two cate-
gories: a) Small organic chemicals databases and b) natural chemicals/compounds
database.

10.3.1.1 Small Organic Chemicals

BindingDB

BindingDB is an open-access, web-accessible database of evaluated binding affini-

ties, aiming principally on the interactions of protein believed to be drug targets with
small, drug-like molecules [21, 22]. The database is accessible at https://www.bindin
gdb.org/bind/index.jsp. As of June 21, 2022, it consists of 41,296 entries containing
2,519,702 binding data for 1,080,101 small molecules and 8810 protein targets. There
are 5988 protein–ligand crystal structures with BindingDB affinity measurements for
274 S. Kar and J. Leszczynski

proteins with 100% sequence identity, and 11,442 crystal structures allowing proteins
to 85% sequence identity. BindingDB also presents the users a BLAST search page
which allows input for an amino acid or nucleic acid sequence. Users can search the
database based on target and compound. The user can refine their search for a target
by giving ranges for experimental data collected for inhibitors like Ki, KD, EC50,
∆G, and pH. The database includes data extracted from the PubChemBioAssays,
scientific reports, and ChEMBL entries for established targets.

ChEBI

ChEBI stands for ‘Chemical Entities of Biological Interest,’ a freely available

database of ‘small molecular entities’ (any constitutionally or isotopically distinct
molecule, atom, ion, ion pair, radical ion, radical, conformer, complex, etc.,), devel-
oped at the EBI [23]. ChEBI is referenced repository for molecular entities and their
ontology based on small chemicals. ChEMBI is accessible at https://www.ebi.ac.uk/
chebi/init.do. ChEBI is available from the EBI FTP site at https://ftp.ebi.ac.uk/pub/
databases/chebi/. ChEBI can be downloaded in the following formats: SDF files,
ontology files. Tools such as TopBraid, OWL-API, the NeOn toolkit and Protégé can
be used with this ontology.

ChEMBL

ChEMBL is another prominent manually curated database of bioactive compounds

with drug-like properties. The database contains chemical, bioactivity, and genomic
data to aid the translation of genomic information into effective new drugs [24, 25].
ChEMBL is freely accessible at https://www.ebi.ac.uk/chembl/. The current release
ChEMBL 30 contains 14,855 targets, 2,157,379 distinct compounds, around 14,000
drugs, 2000+ cells, 752 tissues, 6300+ mechanism, 19,286,751 activities, 84,092
publications as literature resources, and 194 deposited datasets. The database is
available at https://www.ebi.ac.uk/chembl/ The ChEMBL offers multiple associated
resources and tools which can be found in Table 10.1.

ChemDB

ChemDB is an open-source chemical database and consists of around 5 M commer-

cially available small molecules [26]. The dataset is publicly accessible at http://
cdb.ics.uci.edu/. These small molecules can be used as probes in systems biology,
synthetic building blocks, and as leads for the drug discovery. Along with chemicals
structure, it includes predicted or experimentally determined physicochemical prop-
erties, optimization of chemical structure. ChemDB primarily contains two types of
datasets, namely chemical datasets and chemical reactivities datasets. The database
includes a text-based search engine based on fuzzy text matching for searching of
10 Databases for Drug Discovery and Development 275

Table 10.1 Resources and tools under ChEMBL

Resource/tools Description
ADME SARfari Tool for prediction and comparison of ADME targets
ChEBML-NTD Primary screening and medicinal chemistry data of neglected tropical
diseases. ChEMBL-NTD is maintained by EMBL-EBI at Hinxton in the
United Kingdom
GPCR SARfari Chemogenomics workbench for G protein-coupled receptor (GPCR)
Kinase SARfari Chemogenomics workbench for kinases incorporating and linking kinase
sequence
Malaria data Compounds, targets, assays, and data for the malaria-related study
SureChEMBL Publicly available database for patent information extracted from multiple
patents documents and authorities
SARS-CoV-2 data Contains 37 K activities, 8200+ compounds, 57 assays and 10 literature
source
UniChem UniChem is large-scale non-redundant database of pointers between
chemical structures and EMBL-EBI chemistry resources. The
cross-referencing between identifiers from different chemical databases

compounds based on over 65 M annotations from over 150 vendors. The built-
in reaction models support searches through virtual chemical space, comprising
of hypothetical compounds instantly synthesizable from the building blocks in
ChemDB.

ChemSpider

ChemSpider is a publicly accessible chemical structure database https://www.che

mspider.com/ offering structure search access to over 114 million chemical structures
from 272 data sources [27]. Users can search chemical names based on systematic
names, synonyms, trade names, and database identifiers. While chemical structure
can be searched based on structure-based queries, drawing structure in the webserver,
and using structure files from the computer, ChemSpider is maintained by the Royal
Society of Chemistry.

Ligand Expo

Ligand Expo, previously known as Ligand Depot offers chemical and structural
information of small molecules within the structure entries of the Protein Data Bank
(PDB) [28]. The database provides tools to search the PDB for chemical components
followed by identification of structure entries containing particular small molecules.
Users can draw the new chemicals under sketch tool. The data is updated weekly.
Access the data freely at http://ligand-depot.rutgers.edu/index.html. Data is available
as per following formats: mmCIF, SDF/MOL, PDBML.
276 S. Kar and J. Leszczynski

NCI Databases

National Cancer Institute databases [29] contain around 90 datasets. The database
consists of 24 clinical, 23 epidemiological, 19 genomic, 14 imaging, 3 biological
networks, and 3 patient registries databases. The databases are freely available under
the following link: https://www.cancer.gov/research/resources/search?from=0&too
lSubtypes=clinical_data&toolTypes=datasets_databases. Databases covering multi-
tude information related to cancer treatment, biology, omics, screening and detec-
tion, cause and diagnosis, health disparities, prevention, public health, and overall
statistics. Along with cancer, many datasets contain AIDS-related data too. For the
information purpose, we have depicted clinical datasets and tools only, in Table 10.2
as this section is specific to chemical data only.

PubChem

PubChem is an open chemistry database at the National Institutes of Health (NIH)

and maintained by National Center for Biotechnology Information (NCBI) [30]. One
can freely access the data as well as deposit their scientific data which other users
may use. PubChem is one of the most prominent resources for researchers, students,
and scientists to avail the chemical information, structure of chemical compounds
in diverse form like 2D/3D structure, 3D conformer, crystal structure information,
IUPAC name/InChI/InChI key/Canonical SMILES along with multiple identifiers.
Pubchem offers chemical and physical properties, spectral information, pharma-
cology and biochemistry, toxicity followed by multitude information associated with
disorders, diseases, and biomolecular interactions and pathways. Users can access
the PubChem site at https://pubchem.ncbi.nlm.nih.gov/. As per June 21, 2022, the
PubChem data counts have been demonstrated in Table 10.3.

SuperLigands

SuperLigands is open-source database of ligand structures obtained from the PDB

[31]. It combines knowledge about drug-likeness and binding properties of small
molecules or ligands. The database provides ligands in the MDL Mol file format
instead of the PDB format. Structural similarity can be estimated through the calcula-
tion of Tanimoto coefficients and by 3D superposition, while 2D similarity search can
be performed by fingerprints. The database is an excellent source for drug discovery
research.

Toxin and Toxin-Target Database (T3DB)

The T3DB (will be referred as the Toxic Exposome Database) is a distinctive source
of toxin data with complete toxin target information [32]. The database consists
of 3678 toxins including pollutants, pesticides, drugs, and food toxins, which are
10 Databases for Drug Discovery and Development 277

Table 10.2 Clinical sub-databases and tools under NCI databases

Databases name Description
AIDS Antiviral Screen Data Checked tens of thousands of compounds for evidence of
anti-HIV activity by the Developmental Therapeutics
Program (DTP)
Cancer Data Access System Submission and tracking system for data from the Prostate,
(CDAS) the National Lung Screening Trial (NLST), Lung, Colorectal
and Ovarian (PLCO) Cancer Screening Trial, and the
Interactive Diet and Activity Tracking in AARP
CLAMP-Cancer A tool used to build customized natural language processing
pipelines for extracting cancer information from pathology
reports through a user-friendly interface
Chemical Data Compound sensitivity data for the NCI60 screen and similar
screens run on small cell lung cancer cell lines and sarcoma
cell lines, plus molecular target characterization data for the
NCI60, sarcoma, and SCLC cell lines
Clinical Proteomic Tumor CPTAC analyzes cancer biospecimens by mass
Analysis Consortium (CPTAC) spectrometry, characterizing and quantifying their
Data Portal constituent proteins, or proteome
Compound Inhibition Bulk Data Contains compound sensitivity data for the NCI60 screen
and similar screens run on sarcoma cell lines and small cell
lung cancer cell lines
Extensible Neuroimaging Imaging informatics platform designed to support
Archive Toolkit (XNAT) institutional image repositories, image-based clinical trials,
and translational imaging research
Genomic Data Commons (GDC) A data sharing platform that offers harmonized genomic and
de-identified clinical data from various large-scale cancer
studies, along with tools for visualization and data analysis
Human Cancer Models Initiative HCMI has two sections. Consent template and searchable
(HCMI) catalog. Informed Consent Template is used for tissue
accrual for cancer model development. A searchable catalog
is for next-generation cancer models and associated clinical
and molecular data
ivyGlimpse A bioconductor-based interface to Ivy Glioblastoma Atlas
Project (Ivy-GAP) data resources, allowing interactive
selection of image features for scatter plotting, image sets for
stratified survival distribution estimation, and gene sets for
expression distribution comparison between strata
Molecular Target Data The NCI60 lab will provide NCI60 cell line frozen cell
pellets, DNA, or RNA for analysis in the external
researchers’ labs
NCI Brain Neoplasia Data It integrates clinical and functional genomics data from
clinical trials involving brain tumor patients and provides the
ability to perform ad hoc querying, reporting, and analysis
across multiple data domains, including gene expression,
gene copy number, and clinical data
(continued)
278 S. Kar and J. Leszczynski

Table 10.2 (continued)

Databases name Description
Patient-Derived Xenograft Tool for integrating, archiving, and disseminating
(PDX) Finder information about PDX models and their associated data
TP53 Database Contains TP53 variants, including functional/structural data,
germline variants, somatic variants, cell lines, mouse
models, and experimentally induced variants
Vizome A data browser used to access molecular characterization
and drug response studies of clinically annotated adult acute
myeloid leukemia cases
Yeast Anticancer Drug Screen Compound sensitivity data for the NCI60 screen and similar
screens run on sarcoma cell lines and small cell lung cancer
cell lines, plus molecular target characterization data for the
NCI60, sarcoma, and SCLC cell lines

Table 10.3 PubChem data counts as per June 21, 2022

Data type Counts Description
Bioactivities 294,881,644 Biological activity data points reported in PubChem BioAssays
BioAssays 1,465,993 Biological experiments provided by PubChem contributors
Compounds 111,451,641 Unique chemical structures extracted from contributed PubChem
Substance records
Data sources 862 Organizations contributing data to PubChem
Genes 103,628 Genes tested in PubChem BioAssays and those involved in
PubChem Pathways and identified in PubChem Patents
Literature 34,208,642 Scientific publications with links in PubChem
Patents 41,796,860 Patents with links in PubChem
Pathways 238,908 Interactions between chemicals, genes, and proteins
Proteins 185,202 Proteins tested in PubChem BioAssays and those involved in
PubChem Pathways and identified in PubChem Patents
Substances 279,279,532 Information about chemical entities provided by PubChem
contributors
Taxonomy 112,603 Organisms of proteins/genes tested in PubChem BioAssays and
those involved in PubChem Pathways and identified in PubChem
Patents

linked to 2073 corresponding toxin targets. Totally, there are 42,374 toxin, toxin
target associations. Each toxin record refers as ToxCard which comprises over 90
data fields and includes data like chemical properties and descriptors, cellular and
molecular interactions, toxicity values, and medical information. The database is
freely accessible at http://www.t3db.ca/ where records, data, structure and protein/
gene sequences can be downloaded as XML, CSV/JASON, SDF and FASTA formats,
respectively. The major aim of the T3DB is to offer precise mechanisms of toxicity
and target proteins for every single toxin. T3DB is modelled after and carefully linked
10 Databases for Drug Discovery and Development 279

Table 10.4 Resources available in the ZINC15 database

Resource Contents Approx. counts
Activities Best ligand-gene affinity 220,000
Atc codes Atc codes 1500
Catalogs Vendor and annotated catalogs 400
Cat items What vendors and annotated catalogs call the molecules in 1 billion
their source catalogs
ecfp4s Lala 100,000,000
Gene relations Gene_relations 250,000
Genes UniProt Gene Symbols 2800
Major classes Major classes 15
Observations Individual reports of ligand-gene associations 280,000
Organisms asdf 5
Orthologs UniProt accession codes, thus species specific 3800
Patterns SMARTS patterns 535 patterns,
2.5 M entries
Predictions SEA predictions 1 billion
Protomers 3D representations 6 million+
Rings Ring systems 10,000
Subclasses Subclasses 44
Substances Molecules 200,000,000
Tool Tool compounds 3000
compounds

to the DrugBank and Human Metabolome Database (HMDB). Prospective purposes

of T3DB consist of toxin/drug interaction prediction, toxin metabolism prediction,
and general toxin hazard alertness by the people, creating it relevant to numerous
fields.

Zinc
ZINC is an open-access database of commercially available compounds for virtual
screening approach. ZINC is presently known as ZINC15 which contains over 230
million purchasable compounds in ready-to-dock accessible in 3D formats and also
includes over 750 million purchasable compounds [33]. The database is acces-
sible at https://zinc15.docking.org/. ZINC is created and maintained by the Irwin
and Shoichet Laboratories in the Department of Pharmaceutical Chemistry at the
University of California, San Francisco (UCSF). Users can freely download the data
in different file formats, including mol2, SMILES, 3D SDF, and DOCK flexibase
format. Searching, browsing, and molecular drawing interface facility are available
on the ZINC database. ZINC15 is currently supporting multiple resources which are
depicted in Table 10.4.
280 S. Kar and J. Leszczynski

10.3.1.2 Natural Compounds Database

BIAdb

BIAdb is a compilation of benzylisoquinoline alkaloids (BIAs) database that stores

information of around 846 unique BIAs where 196 entries from KEGG, 145 data
from CTD, 171 entries from 171 and 334 data from other literature source [34]. The
database can be downloaded from the following website at https://webs.iiitd.edu.in/
raghava/biadb/. As BIAs have therapeutic properties, they can be a good resource
for virtual screening to obtain potential lead molecules. Accessible natural alkaloids
are produced by a range of organisms, like fungi, bacteria plants, and animals. The
entire list of alkaloids underBIAdb is depicted in Table 10.5.

Dictionary of Natural Products Online

The Dictionary of Natural Products is an online resource for natural products [35].
It is resulting from a Dictionary of Organic Compounds (DOC), a repository of
natural product and the data has been accumulated by a team at Chapman and Hall,
UK. Comparable compounds are coordinated into a single entry streamlining the
relationships of those strongly associated compounds. Compounds are indexed by
their structural and biogenetic type. The dictionary is equipped with advanced search
option with different properties like melting point and boiling point along with CAS,
chemical and molecular formula.

Naturally Occurring Plant-Based Anticancer Compound-Activity-Target

Database (NPACT)

NPACT is a compilation of plant-derived natural compounds showing anticancer

activity under in vitro and in vivo experiments [36]. The present version contains
around 1574 compound entries. The database offers chemical structure, data on
in vitro and in vivo experiments along with inhibitory data like ED50 /IC50 /GI50 /EC50
and physical, topological and elemental properties. User have also access to drug-
likeness, target information, cancer types, references, and vendors information of
respective compounds. NPACT can be a great starting point in the drug discovery of
cancer.

SuperNatural II

SuperNatural II is an open-access database for natural products. It offers 325,508

natural compounds with information about the 2D structures, corresponding physic-
ochemical properties, and predicted toxicity data [8]. Extreme chemically diverse
natural products give enormous prospect for researchers to innovate new drug
10 Databases for Drug Discovery and Development 281

Table 10.5 Types of alkaloids under BIAdb

Chemical class Sub-class Alkaloids name
Indole Beta-carbolines Harmine, harmaline, tetrahydroharmine
Ergolines Ergine, ergotamine, lysergic acid
Mitragyna Mitragynine, 7-hydroxymitragynine
speciosa
Strychnos Strychnine, brucine
nux-vomica
Tabernanthe iboga Ibogaine, voacangine, coronaridine
Tryptamines Serotonin, DMT, 5-MeO-DMT, bufotenine, psilocybin
Vinca Vinblastine, vincristine
Yohimbans Reserpine, yohimbine
Isoquinoline Opium Papaverine, narcotine, narceine
– Sanguinarine, hydrastine, berberine, emetine,
berbamine, oxyacanthine
Phenanthrene Opium Morphine, codeine, thebaine
Phenethylamine – Mescaline, ephedrine, dopamine
Purine Xanthines Caffeine, theobromine, theophylline
Pyridine – Piperine, coniine
Pyrrolidine – Hygrine, cuscohygrine, nicotine
Quaternary Muscarine, choline, neurine
ammonium
Quinoline – Quinine, quinidine, dihydroquinine, dihydroquinidine,
strychnine, brucine, veratrine, cevadine
Terpenoid Aconite Aconitine
Steroid Solanum alkaloids: solanidine, solanine, chaconine;
Veratrum: veratramine, cyclopamine, cycloposine,
jervine, muldamine; newt: samandarin
Tropane – Atropine, cocaine, ecgonine, scopolamine, catuabine
Miscellaneous – Capsaicin, cynarin, phytolaccine, phytolaccotoxin

discovery, nutritional products, cosmetics, and agrochemical research. The database

is accessible at https://bioinf-applied.charite.de/supernatural_new/index.php?site=
home. Users can search natural compounds based on properties, by name or by
providing templates like amino acids, alpha sugars, D-sugars, aromatics, bases,
bicycles, fused rings, heterocyclic rings.
282 S. Kar and J. Leszczynski

10.3.2 Drug Molecules Database

10.3.2.1 DrugBank

DrugBank is a comprehensive, open-access, online database and consists of infor-

mation on drugs and drug targets [11]. All drugs can be downloaded from
https://go.drugbank.com/releases/latest. DrugBank is one of the most popular
databases for drug design and discovery which contains detailed knowledge about
drugs’ chemical, pharmacological, and pharmaceutical data. The latest version of
DrugBank (version 5.1.10, released 2023-01-04) consists of 15,321 drugs including
2734 approved small molecule drugs, 1572 approved biologics (peptides, proteins,
vaccines, allergenics), 134 nutraceuticals, and over 6716 experimental (discovery-
phase) drugs. Furthermore, 5294 non-redundant protein (i.e., drug target/enzyme/
transporter/carrier) sequences are linked to these drug entries. Each item includes
over 200 data fields with half of the information being devoted to drug/chemical
data and the other half devoted to drug target or protein data. DrugBank also
offers information related to pharmacological pathways, drug reactions, pharmacoge-
nomics, metabolomics, transcriptomics, and proteomics. In recent time, DrugBank
has created special dashboard for COVID-19-related drug information.

10.3.2.2 PharmGKB

Pharmacogenomics Knowledgebase (PharmGKB) offers data on the effect of human

genetic variation on drug responses. PharmGKB is created by the NIH and managed at
Stanford University. It is one of the partners of the NIH Pharmacogenomics Research
Network (PGRN). The database contains clinical data, pharmacokinetics, and phar-
macogenomics data in pulmonary, cancer, cardiovascular, and metabolic pathways
domains. The database offers data on 832 drug label annotations, 201 curated path-
ways, 188 clinical guideline annotations, and 746 annotated drugs [37]. The drug
labels comprising pharmacogenetic information approved by the US FDA, Swiss
Agency of Therapeutic Products (Swissmedic), European Medicines Agency (EMA),
Health Canada (Santé Canada) (HCSC), and Pharmaceuticals and Medical Devices
Agency Japan (PMDA). All the available resources can be downloaded at free of
charge from https://www.pharmgkb.org/downloads.

10.3.2.3 SuperDRUG2

SuperDRUG2 is a comprehensive knowledgebase of approved and marketed drugs

[12]. The database offers around 4600 active pharmaceutical ingredients as per last
release 2018.2.7. SuperDRUG2 annotated drugs with regulatory information, chem-
ical structures, physicochemical properties, dosage, biological targets, side effects,
and pharmacokinetic data. Users can search the chemical space of approved drugs
10 Databases for Drug Discovery and Development 283

through a different mechanism. It also offers a 2D chemical structure search on top of

a 3D superposition feature that superposes a drug with ligands found in the protein–
ligand complexes. The interaction check feature can detect possible drug–drug inter-
actions, which includes alternate suggestions for geriatric patients. SuperDRUG2 can
be accessed freely for academia and requires a free browser plugin called “Chime”
for visualization.

10.3.3 Therapeutic Target Database

Proteins, enzymes, and nucleic acids are potential therapeutic targets for diseases.
Therefore, binding interaction of small drug molecules to macromolecules like
protein and/or protein–protein interactions is significant to understand developing
new drug candidates for a specific disease. Understanding protein’s structure and
functions is essential to understand the pharmacological mechanism of small
molecules binding to a specific protein. A therapeutic target database can offer more
detailed information about different drug design and discovery targets. The most
commonly employed therapeutic target databases are discussed below.

10.3.3.1 Herbal Ingredients Targets Database (HIT) 2.0

HIT 2.0 is developed based on the most updated curated database focusing on Herbal
Ingredients’ Protein Targets covering PubMed literature 2000–2020 and precursors
for FDA-approved drugs [38]. HIT 2.0 hosts 10,031 ingredient-target activity pairs
with quality indicators between 2208 biological targets and 1237 herbal ingredients
from 1250 source herbs. The database has also consisted of 1231 therapeutic targets
and 56 micro RNA targets. The molecular targets cover those genes/proteins that
are directly/indirectly activated/inhibited, protein binders, and enzyme substrates or
products. Also included are those genes regulated under the treatment of individual
ingredients. HIT can be freely reachable at http://hit2.badd-cao.net.
HIT facilitates automated target-mining and My-target curation, where users can
retrieve and download the latest abstracts containing potential targets for concerning
herbs. The database contains molecular target information, which encompasses those
proteins being activated or inhibited, protein binders, and enzymes whose substrates
or products are interesting compounds. On the other hand, users can enter the ‘My-
target’ curation system to curate the comprehensive ingredient-target relationship
and create the latest individual targeting profiles.

10.3.3.2 Molecular Modeling Database (MMDB)

The MMDB is an open-access database that contains experimentally determined

3D biological macromolecular structures, including proteins and polynucleotides.
284 S. Kar and J. Leszczynski

The database is maintained by the National Center for Biotechnology Informa-

tion (NCBI), USA. MMDB can be freely accessed at http://www.ncbi.nlm.nih.
gov/structure. It is linked to NCBI’s Entrez search and retrieval systems, including
contents of protein structures of PDB, PubMed, nucleotide and protein sequences,
taxonomy, complete genomes, etc. [39]. MMDB offers accurate and pre-computed
structural alignments obtained by the Vector Alignment Search Tool (VAST)
(accessed at: http://www.ncbi.nlm.nih.gov/Structure/VAST/vastsearch.html) and
also provides visualization tools for 3D structure and sequence alignment with
molecular graphics tool Cn3D (available at: http://www.ncbi.nlm.nih.gov/Structure/
CN3D/cn3d.shtml. CBLAST (accessed at: http://www.ncbi.nlm.nih.gov/Structure/
cblast/cblast.cgi) is another web service that visualizes similarities between proteins
in NCBI’s Entrez database and those with known 3D structures tracked in MMDB.

10.3.3.3 Protein Data Bank (PDB)

The PDB is a freely accessible (https://www.rcsb.org/) archive for the three-

dimensional (3D) structures of biological macromolecules like proteins, nucleic
acids, and complex assemblies [13]. The obtained data is created through 3D structure
elucidation techniques such as X-ray crystallography, NMR spectroscopy, or cryo-
electron microscopy. Users can get the 3D structure of proteins and their complexes
with other molecules with different resolutions, organisms, and expression systems
at PDB. For docking and molecular dynamics study, PDB is the most commonly
accessible site to obtain the target for structural biology and drug designing. Users
cannot only access the X-ray crustal structure of PBD but also deposit their analyzed
X-ray crystal structure. The PDB database consists of advanced ‘Search,’ ‘Visualize,’
and ‘Analyze’ tabs with many scientific analysis options. One can avail of the PDB
site from the multiple websites of its member organizations like PDBe, RCSB, and
PDBj. The PDB archive is maintained and organized by Worldwide Protein Data
Bank (wwPDB) [40].

10.3.3.4 Therapeutic Target Database (TTD)

TTD is a database about the known therapeutic protein, nucleic acid targets, the
targeted disease, pathway information, and the corresponding drug targets [41]. The
database is open-access at http://db.idrblab.net/ttd/. The database is created and main-
tained by constructed by the Innovative Drug Research and Bioinformatics Group
(IDRB) (Zhejiang University, China) and the Bioinformatics and Drug Design Group
(BIDD) (National University of Singapore). Users can search the database based on
‘Search for drugs,’ ‘Search Drugs and Targets by Disease or ICD Identifier,’ ‘Search
for Biomarkers,’ and ‘Search for Drug Scaffolds.‘ TTD is well referenced to related
databases accumulating information on sequence, target function, 3D structures, drug
structure, therapeutic class, enzyme nomenclature, ligand properties, and category
of clinical development. Multi-target agents have been studied to improve the safety
10 Databases for Drug Discovery and Development 285

profiles, therapeutic activity, and resistance by modulating the activity of a primary

target. The recently updated database includes 1308 targets with 12,683 non-binders
and 34,861 poor binders; (1) 1127 co-targets of 672 targets regulated by 642 approved
and 624 clinical trial drugs; (2) 534 prodrug-drug pairs for 121 targets; (3) the profiles
of drug-like properties of 33,598 agents’ of 1102 targets; and (4) structure–activity
landscapes of 427,262 active agents of 1565 targets [41]. The database offers further
data and function, including cross-links to the target structure in PDB and AlphaFold,
159 and 1658 newly emerged targets and drugs.

10.3.3.5 Universal Protein Resource (UniProt)

The UniProt consortium is created by the European Bioinformatics Institute

(EBI), the Swiss Institute of Bioinformatics (SIB), and the Protein Information
Resource (PIR) [42]. It offers data on protein sequences and their functions where
entries originated from genome sequencing projects. UniProt is freely accessible at
https://www.uniprot.org/. The database for protein sequences is specifically known
as UniProtKnowledgeBase (UniProtKB), which has two sections named UniProtKB/
Swiss-Prot and UniProtKB/TrEMBL. The first consists of manually annotated entries
curated from the literature and curator-evaluated computational analysis. In contrast,
the latter one stores computer-annotated entries, which await complete manual anno-
tation. In the UniProt database, sequence clusters, sequence archives, and proteome
sets are available under UniRef, UniParc, and Proteomes tabs.
1. UniRef: UniRef offer clustered sets of sequences from the UniProtKnowledge-
Base and selected UniParc records. UniRef hides superfluous sequences and gets
comprehensive coverage of the sequence space at three different resolutions.
UniRef100 contains identical sequences and sub-fragments with 11 or more
residues from any organism into a single UniRef entry. UniRef90 is prepared by
clustering UniRef100 sequences that have at least 90% sequence identity and 80%
overlap with the longest sequence. UniRef50 is created by clustering UniRef90
seed sequences that have at least 50% sequence identity and 80% overlap with
the longest sequence in the cluster.
2. UniParc: UniParc is a complete and non-redundant database that contains
publicly available protein sequences. UniParc avoids duplicate copies of protein
by storing each unique sequence only once and providing it a unique identifier
(UPI). UniParc contains only protein sequences, while cross-referencing retrieves
other information about the protein from the source databases.
3. Proteomes: A proteome is a set of proteins considered to be expressed by an
organism. UniProt proteomes provide proteomes for species with completely
sequenced genomes.
286 S. Kar and J. Leszczynski

10.3.4 Peptide Database

10.3.4.1 Antimicrobial Peptide Database (APD)

The APD contains 3425 antimicrobial peptides (AMP) from six life kingdoms of
natural sources [43]. The antimicrobial activities are demonstrated for the included
peptides, and the activity range is either MIC < 100 µM or 100 µg/ml. The natural
sources are the following: 2489 from animals, including some synthetic peptides, 385
isolated/predicted bacteriocins/peptide antibiotics from bacteria, 368 from plants, 25
from fungi, 8 from protists, and 5 from archaea). A complete list of 3425 AMPs under
APD is listed in Table 10.6.
To avail of the database, one should visit https://aps.unmc.edu/. The database
comprises a pipeline of search functions for innate immune peptides. One can search
for peptide information utilizing APD ID, amino acid sequence, chemical modifica-
tion, peptide name, peptide motif, length, hydrophobic content, charge, 3D structure,
PDB ID, peptide source organism, methods for structural determination, peptide
family name, life domain/kingdom, biological activity, target microbes, synergistic
effects, molecular targets, mechanism of action, and publications details.

10.3.4.2 CAMPR3

The collection of antimicrobial peptides (CAMPR3) is a rich resource for antimi-

crobial peptide family-based information. Antimicrobial peptides are family-specific
sequence compositions that can be used to design and discover novel AMPs [44]. The
CAMPR3 comprises data on the preserved sequence signatures captured as patterns
and hidden Markov models (HMMs) in 1386 AMPs characterized by 45 families.
Information connected to protein definition, sequence, activity, accession numbers,
target organisms, source organism, protein family descriptions is freely available.
The database also offers pattern creation, sequence alignment, AMP prediction,
pattern and HMM-based search tools. The database is also linked to databases like
PubMed, UniProt, and other antimicrobial peptide databases. The site is accessible
at: http://www.camp.bicnirrh.res.in/. A detailed, accessible data statistic is shown in
Table 10.7.

10.3.4.3 CancerPPD

CancerPPD is a database that offers experimentally verified anticancer peptides

(ACPs) and proteins. All the available data was mined manually from peer-reviewed
literature and patents [45]. The database predicted tertiary structures of anticancer
peptides employing the PEPstr method and secondary structure states are assigned
using DSSP. One can browse the database based on protein, peptide, tissue, cell line,
10 Databases for Drug Discovery and Development 287

Table 10.6 Antimicrobial activity, resource statistics of peptides under APD

Function information Count
Antibacterial peptides 2891
Antiviral peptides 200
Antifungal peptides 1252
Anti-candida peptides 721
Antibiofilm peptides 75
Antiparasital peptides 140
Insecticidal peptides 41
Spermicidal peptides 14
Anti-HIV peptides 109
Anticancer (antitumor) peptides 264
Chemotactic peptides 67
Wound healing peptides 25
Antioxidant peptides 30
Enzyme/protease inhibitory peptides 33
Immobilized peptides 31
Anti-MRSA peptides 208
Antitoxin peptides 15
Channel inhibitors 7
Antiinflammatory peptides 32
Antidiabetic peptides 16
Anti-TB peptides 14
Antiendotoxin peptides 88
Two-chain peptides 35
Synergistic peptides 47
TOTAL 3425
Resources Counts
Human host defense peptides 147
Active peptides from amphibians (frogs/toads) 1148 (1070/74)
Fish peptides 140
Reptile peptides 45
Mammals annotated 352
Birds 43
Molluscs 47
Protozoa 6
Insects 339
Crustaceans 73
(continued)
288 S. Kar and J. Leszczynski

Table 10.6 (continued)

Resources Counts
Myriapods 8
Spiders 44
Scorpions 93

Table 10.7 Statistics of the CAMPR3 database

Database Information
types
Sequences Contains 8164 AMP sequences covering taxonomy of algae, amoebozoa,
animalia, archaea, bacteria, fungi, heterolobosea, viridiplantae, virus, synthetic
construct
Structures 757 AMP structures covering activity for antibacterial, antifungal, antiviral,
unclassified
Patents 2083 patented AMPs are available
Signatures 36 patterns and 78 HMMs

assay, etc. The database is available at: http://crdd.osdd.net/raghava/cancerppd/ind

ex.php. The available total peptides, cell lines, and tissue types under CancerPPD
are 3491, 249, and 21, respectively. Major features under this database are (1) data
retrieval where data fetching and advanced search are possible along with search
of peptides; (2) data analysis through multiple tools like BLAST, Smith-waterman,
sequence and structure mapping followed by similarity-based search, (3) availability
of ACPs SMILES and structures, (4) prediction of tertiary structures of all ACPs
are also accessible. The CancerPPD also offers data connected to diverse chemical
modifications like D-amino acids, non-natural, modified-amino acid like ornithine.

10.3.4.4 StraPep

Structure database of bioactive peptides (StraPep) is a dedicated database of bioactive

peptides with known structures [46]. The present version of the database contains
3791 bioactive peptide structures and comprises 1312 unique bioactive peptide
sequences. The StraPep is categorized into six functional groups counting antimi-
crobial peptide (404/833) with the distribution of 30.79%, toxin and venom peptide
(464(unique sequences)/885 (structures)) covering 35.37%, cytokine and growth
factor (217/901) having 16.54%, hormone (141/860) having 10.75%, neuropeptide
(39/60) of 2.97%, and others (47/252) sharing 3.58%. The database is accessible
at: http://isyslab.info/StraPep/. The peptides can be browsed based on classification,
organism, disulfide bond, and cystine knot. The StraPep is also connected with tools
like Blastp, Map, and Secondary Structure Composition search.
10 Databases for Drug Discovery and Development 289

10.3.5 Metabolomic Database

10.3.5.1 Biochemical Genetic and Genomic (BiGG)

BiGG is a knowledgebase of large-scale biochemically, genetically and genomically

structured genome-scale metabolic reconstructions under the constraint-based recon-
struction and analysis (COBRA) framework which are valuable tools for evaluating
the metabolic capacities of organisms and interpretation of experimental data [47].
BiGG is freely available for academic users at http://bigg.ucsd.edu/. BiGG can be
utilized to browse model content, visualize metabolic pathway maps, and export
SBML files of the models for additional assessment. Users may follow links from
BiGG to several external databases to obtain additional information on proteins,
genes, metabolites, reactions, and citations of interest. BiGG contains 9088 metabo-
lites with specific ID and names. This database focuses on the need for systems
biology scientists by delivering 75 genome-scale high-quality metabolic models
under BiGG models [48].

10.3.5.2 BioCyc

BioCyc is a compilation of 20,005 pathway/genome databases (PGDBs) for model

eukaryotes and thousands of microbes and tools for exploring them [49]. BioCyc is
encyclopedic which contains curated data from 130,000 literatures. The BioCyc is
available at https://biocyc.org/ but requires a subscription to use. Under BioCyc, 470
databases are for archaea, 19,416 for bacteria, 37 for eucaryota, and the remaining
are for metabolic databases (named MataCyc). BioCyc incorporates knowledge from
other bioinformatics databases, for instance, protein feature and Gene Ontology infor-
mation from UniProt, gene-essentiality datasets from OGEE, and regulatory informa-
tion from RegTransBase. BioCyc offers a suite of bioinformatics tools like Search
across organisms and databases, visualization, genome browser, omics data anal-
ysis, SmartTable, metabolic route search, comparative analysis, sequence analysis,
and pathway tools software. The database has three tier PGDBs where Tier 1 can be
manually curated and frequently updated and include EcoCyc, HumanCyc, MetaCyc,
AraCyc, YeastCyc, and the BioCyc Open Compounds Database (BOCD). Tier 2 is
generated computationally by PathoLogic program used to predict their metabolic
pathways, persons, and pathways hole filers. It has moderate manual updating. It
contains 64 databases as per the present version. However, Tier 3 has 19,936 databases
which were computationally generated and receive no manual updates.

10.3.5.3 Human Metabolome Database (HMDB)

The HMDB database contains information about small molecule metabolites in the
human body [50]. The database is anticipated to be applied in clinical chemistry,
290 S. Kar and J. Leszczynski

metabolomics, and biomarker discovery. The database has three forms of data: (1)
clinical data, (2) chemical data, and c) molecular biology/biochemistry data. The
database is freely accessible at https://hmdb.ca/. The HMDB database is released
every two years with monthly corrections and updates. The current Version (5.0)
contains 220,945 metabolites and 8610 protein sequences, including enzymes and
transporters linked to these metabolite entries. Metabolite structures are available
in SDF format, protein and gene sequences are in FASTA format, and metabolite
and protein data are in XML format. Individual MetaboCard entry contains 130 data
fields, with two-thirds of the material being dedicated to chemical and clinical data
and the other one-third committed to enzymatic or biochemical data. The HMDB
endorses extensive text, chemical structure, sequence, NMR, and MS spectral query
searches. Databases like T3DB, DrugBank, FooDB, and SMPDB are also part of the
HMDB suite. Metabolites can browse by multiple filter options like metabolite status,
biospecimen (saliva, blood, urine, feces, cerebrospinal fluids, breast milk, bile, sweat,
amniotic fluids), origin (endogenous, exogenous, plant, food, microbials, toxins,
cosmetics, drugs) and cellular location (cell membrane, cytoplasm, mitochondria,
nucleus).

10.3.5.4 Kyoto Encyclopedia of Genes and Genomes (KEGG)

KEGG is an excellent resource for genomes, diseases, biological pathways, drugs,

and chemicals, offering an understanding of high-level functions and utilities of
biological systems [51]. It is a large-scale molecular database created by genome
sequencing and other high-throughput experimental technologies. The current
release of the database can be found at https://www.genome.jp/kegg/. KEGG
offers many information on chemicals, genes, and genomes, followed by health
information. It consists of a set of tools for diverse analysis, which can be found in
Table 10.8.

10.3.5.5 MetaboLights Database

MetaboLights is a database for metabolomics experiments and derived information

[52, 53]. The database is cross-platform, cross-species, cross-technique metabolomic
research performed at the European Bioinformatics Institute (EMBL-EBI). The
database offers experimental data from metabolomics experiments compliant with
Metabolomics Standards Initiative (MSI). The database also offers metabolite struc-
tures, reference spectra, biological roles, locations, and concentrations. It has robust
reporting capabilities and provides user-friendly submission tools. The database is
accessible at https://www.ebi.ac.uk/metabolights/. Users can search data based on
studies, compounds, and species.
10 Databases for Drug Discovery and Development 291

Table 10.8 Information available under KEGG database

Database Narration
Chemical information
COMPOUND A repository of small molecules, biopolymers, and other chemicals relevant to
biological systems
ENZYME Collection of information about enzyme nomenclature (EC number system)
based on ExplorEnz database
GLYCAN A compilation of experimentally determined glycan structures
REACTION A depository of chemical reactions from KEGG metabolic pathway maps
enzyme nomenclature
Genomic information
GENES A collection of gene catalogs from NCBI RefSeq and GenBank. The catalog
contains 41,718,457 genes in KEGG organisms, 595,312 viral genes, 284
viral mature peptides, and 4106 addendum proteins
GENOME A collection of organisms with complete genome sequences and selected
viruses with relevance to diseases
ORTHOLOGY A database of molecular functions as functional orthologs with the present
tally of 25,221
Health information
DRUG Collection of approved drugs in the USA, Europe, and Japan
DISEASE A collection of human disease entries aiming only on the perturbation basis
ENVIRON A collection of health-promoting natural products of plants such as crude
drugs, essential oils, etc
MEDICUS An integrated resource of diseases, drugs, and health-related substances
NETWORK To capture knowledge on diseases and drugs in terms of perturbed molecular
networks
Organism information
Organisms Collection of complete genomes of 762 eukaryotes, 7043 bacteria, and 389
archaea
Systems information
BRITE A collection of hierarchical text (htext) files storing functional hierarchies of
biological objects (KEGG objects). The functional hierarchies and reference
(total) are 185 and 318,231, respectively
MODULE Modules are manually defined functional units identified by the M numbers.
KEGG These are used for annotation of sequenced genomes. The number of KEGG
and reaction modules are 459 and 46, respectively
PATHWAY Contains pathway maps with molecular interaction and reaction. The number
of pathway maps and reference (total) is 551 and 937,937, respectively
Tools for analysis
KEGG mapper KEGG PATHWAY/BRITE/MODULE mapping tools
BlastKOALA BLAST-based KO annotation and KEGG mapping
GhostKOALA GHOSTX-based KO annotation and KEGG mapping
(continued)
292 S. Kar and J. Leszczynski

Table 10.8 (continued)

Database Narration
KofamKOALA HMM profile-based KO annotation and KEGG mapping
BLAST/FASTA Sequence similarity search
SIMCOMP Chemical structure similarity search

10.3.5.6 Small Molecule Pathway Database (SMPDB)

SMPDB is an interactive and visual database comprising more than 30,000 small
molecule pathways found in humans only [54, 55]. The database is unique as most
of the included pathways are not accessible in any other pathway database. The
main aim of this database is to support pathway discovery and its elucidation in
proteomics, metabolomics, transcriptomics, and systems biology. It offers detailed
information about each pathway, fully searchable, hyperlinked diagrams of human
metabolic pathways, metabolite signaling pathways, metabolic disease pathways,
and drug-action pathways. The database is accessible at https://www.smpdb.ca/.
The most recent version is SMPDB v2.75. SMPDB pathways include knowledge
of related organs, subcellular compartments, protein_complex locations, protein_
complex cofactors, protein_complex quaternary structures, chemical structures, and
metabolite locations. Gene, metabolite, and protein_complex concentration data can
also be visualized through SMPDB’s mapping interface. SMPDB’s images, image
maps, descriptions, and tables are downloadable.

10.3.5.7 WikiPathways

WikiPathways is an open-accessed collective platform for acquiring and distributing

models of biological pathways for data visualization and analysis [56]. WikiPathways
is accessible at http://www.wikipathways.org. It offers services to support pathway
analysis and visualization via popular standalone tools, like PathVisio and Cytoscape,
web applications, and standard programming environments. WikiPathways plat-
form is also open to community participation (https://github.com/wikipathways).
WikiPathways comprises over 2300 pathways across over 25 different species. The
human pathway compilation is the biggest and most active collection by species,
having expanded sixfold to include 640 pathways. In terms of coverage of unique
human genes, WikiPathways is comparable to KEGG. Additionally, it consists of
more than 640 pathways from humans encompassing more than 7500 genes and
stores pathways with more than 1000 metabolites.
10 Databases for Drug Discovery and Development 293

10.4 How to Select the Database for the Research?

Databases are critical resources for drug discovery using virtual screening and drug
repurposing approaches. Most of the time, the researcher faces difficulty choosing
the correct database for their research work. Therefore, a thorough understanding
of the database’s nature, type of information and data under the database, source of
the data, accessibility of data in the form of download, as well as open-source or
commercial nature needs to be understood. All drug databases have some pros and
cons. One cannot mention that any specific database is perfect; each is unique on its
own. We have identified five significant characteristics or features summarized in the
user-friendliness score. The highest score is five (all five characteristics are present
for an ideal database), and the lowest score is zero (no characteristic is present). The
characteristics are discussed below:
1. Updated: The databases should have a registered website with regular updates.
In many cases, we have seen after the first release of the database, they are not
updated or maintained for several years. In many cases, even they have stopped
the update. As data is changing daily, data updates and all respective information
related to data need to be available and updated.
2. Advanced search option: Each researcher’s requirement differs from one
another. Therefore, an advanced search option with filters is one of the signif-
icant criteria. DrugBank is one of the ideal databases for the advanced search
option where users can perform a multidimensional search based on chemical
information, pharmacological aspects, metabolism, etc.
3. Downloadable: This is one of the significant characteristics as users need to
use the database for further analysis, for example, virtual screening. Now, if a
database is not downloadable and exportable in major acceptable formats, then
the database is useless. Instead, the database is merely information provided
without further user analysis.
4. Classified: The drug databases are large and contain multitude of informa-
tion. Thus, classifying them into multiple categories can assist the researcher
in obtaining the necessary data with a proper approach. Classification not only
categorizes the data but also helps to narrow down the requirement of the
researchers.
5. Accessibility: Drug discovery is a time-consuming and expensive process. Thus,
in many cases, academics and independent researchers cannot have access to
many commercial databases. They must rely on open-access/freely available
drug databases for their research. Therefore, open-access databases not only
help the researcher to gather enormous resources in no time without spending
money. Once they know which drug they have to look for experimental analysis,
they can only buy the small molecule/peptide directly from the supplier, either
connected with the databases or from external resources.
Considering the discussed characteristics, major databases are plotted using a
user-friendliness score in Fig. 10.5, which shows that CheMBL, DrugR+, DTC,
294 S. Kar and J. Leszczynski

Fig. 10.5 User-friendliness of different drug databases. * Full form of the databases can be found
in the abbreviation section

KEGG, TDR, and TTD scored 5. Here, it is essential to mention that we have only
classified and scored based on our denoted features. Other databases also can be a
better choice depending on the researcher’s requirements. The main idea is to show
how to choose databases for drug discovery only, not to identify the best database.

10.5 Overview and Conclusion

Computational drug discovery research is growing daily, with high-performing

computational resources available. Therefore, with minimal time, the researcher can
screen million to billion of compounds based on their studied targets and diseases.
Drug repurposing is another essential method where an existing USFDA drug for
a specific disease can be used to cure another disease with the help of ligand- and
structure-based drug design and screening approaches. We have seen the use of
remdesivir, a nucleotide analog prodrug initially developed for the treatment of the
Ebola virus, was observed to inhibit the replication of coronaviruses in vitro and in
preclinical studies. This classic example helps us understand that proper screening
of existing approved drugs, drugs under clinical trials, drugs under investigation,
drugs under experimental state, etc., can be beneficial to finding the cure for many
diseases. In this perspective, this enormous number of databases consisting of small
drug molecules, peptides, metabolites, and natural products can be used for the future
mining of potential drug candidates for rare and neglected diseases with minimal time
10 Databases for Drug Discovery and Development 295

and money. Futuristically, integrating computational tools and databases can make
the computer-aided drug design process more realistic as all the resources can be
found in a single place and are more accessible. Another essential aspect that must
be addressed for existing and future databases is the transparency of the data so that
users can have high confidence to employ the data in their research.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this chapter.

Acknowledgements SK wants to thank the administration of Dorothy and George Hennings

College of Science, Mathematics and Technology (HCSMT) of Kean University for providing
research opportunities and resources.

Abbreviations

CHM CheMBL
CHSP Anticancer Herbs database for System Pharmacology;
CMAP Complement Map Database
D2G Drug to Gene
DHUB Drug Repurposing Hub
DMAP Drug-Protein Connectivity MAP
DMC Drug Map Central
DNET Drug-Disease Network Database
DPTH Drug Pathway Database
DRAR Drug Repurposing Adverse Reaction
DrugB DrugBank
DSDB Drug Signatures Database
DSRV Drug Survival Database
DTC Drug Target Commons
DTOM Drug Target Interactome Database
DTW Drug Target Web
GSDB Gene Set Database
HIVRT HIV Drug Resistance Database
KEGG Kyoto Encyclopedia of Genes and Genomes
KSRPO A Platform for Drug Repositioning
NNFIN Network-based Similarity Finder
ODB Ontario Database
PDTD Potential Drug Target Database
PROM Promiscuous
SBIOS Swiss BIOisostere
SCYP Super Cytochrome P450
SIDER Side Effect Resource
296 S. Kar and J. Leszczynski

SUT SuperTarget database

TBDB Tuberculosis Database
TCM Traditional Chicness Medicine
TCMSP Traditional Chinese Medicine Platform
TDR Tropical Diseases Research
THIN The Health Improvement Network
TTD Therapeutic Target Database

References

1. Medina-Franco JL (2021) Grand challenges of computer-aided drug design: the road ahead.
Front Drug Discov 1:728551
2. Mohs RC, Greig NH (2017) Drug discovery and development: Role of basic biological research.
Alzheimers Dement 3:651–657
3. Wouters OJ, McKee M, Luyten J (2020) Estimated research and development investment
needed to bring a new medicine to market, 2009–2018. JAMA 323:844–853
4. Jumper J, Evans R, Pritzel A et al (2021) Highly accurate protein structure prediction with
AlphaFold. Nature 596:583–589
5. Tang Y, Zhu W, Chen K, Jiang H (2006) New technologies in computer-aided drug design:
toward target identification and new chemical entity discovery. Drug Discov Today: Technol
3:307–313
6. Potemkin V, Potemkin A, Grishina M (2018) Internet resources for drug discovery and design.
Curr Top Med Chem 18:1955–1975
7. Miller M (2002) Chemical database techniques in drug discovery. Nat Rev Drug Discov 1:220–
227
8. Banerjee P, Erehman J, Gohlke BO, et al. (2015) Super natural II—A database of natural
products. Nucleic Acids Res 43(Database):D935–D939
9. Gosh S, Kar S, Leszczynski J (2020) Ecotoxicity databases for QSAR modeling. In: Roy K
(ed) Ecotoxicological QSARs. Humana, New York, pp 709–758
10. Kumar V, Roy K (2020) Development of a simple, interpretable and easily transferable QSAR
model for quick screening antiviral databases in search of novel 3Clike protease (3CLpro)
enzyme inhibitors against SARS-CoV diseases. SAR QSAR Env Res 31:511–526
11. Wishart DS, Feunang YD, Guo AC et al (2018) DrugBank 5.0: a major update to the DrugBank
database for 2018. Nucleic Acids Res 46:D1074–D1082
12. Siramshetty VB, Eckert OA, Gohlke BO et al (2018) SuperDRUG2: a one stop resource for
approved/marketed drugs. Nucleic Acids Res 46:D1137–D1143
13. Berman HM, Westbrook J, Feng Z et al (2000) The protein data bank. Nucleic Acids Res
28:235–242
14. Roy K, Kar S, Das RN (2015) Understanding the basics of QSAR for applications in
pharmaceutical sciences and risk assessment. Academic Press
15. Kar S, Leszczynski L (2021) QSAR and machine learning modeling of toxicity of nanoma-
terials: a risk assessment approach. In: Njuguna J, Pielichowski K, Zhu H (eds) Health and
environmental safety of nanomaterials. Woodhead Publishing, pp 417–441
16. Ojha PK, Mitra I, Kar S, Das RN, Roy K (2012) Lead hopping for PfDHODH inhibitors as
antimalarials based on pharmacophore mapping, molecular docking and comparative binding
energy analysis (COMBINE): a three-layered virtual screening approach. Mol Inform 31:711–
718
17. Kumar V, Kar S, De P, Roy K, Leszczynski J (2022) Identification of potential antivirals against
3CLpro enzyme for the treatment of SARS-CoV-2: a multistep virtual screening study. SAR
QSAR Env Res 33:357–386
10 Databases for Drug Discovery and Development 297

18. Kar S, Roy K (2013) Prediction of milk/plasma concentration ratios of drugs and environmental
pollutants using in silico tools: classification and regression based QSARs and pharmacophore
mapping. Mol Inform 32:693–705
19. Kar S, Leszczynski L (2020) Open access in silico tools to predict the ADMET profiling of
drug candidates. Expert Opin Drug Discov 15:1473–1487
20. Kar S, Roy K, Leszczynski L (2020) In silico tools and software to predict ADMET of new
drug candidates. In: Benfenati E (ed) In Silico Methods for Predicting Drug Toxicity. Humana,
New York, pp 85–115
21. Gilson MK, Liu T, Baitaluk M, Nicola G, Hwang L, Chong J (2016) BindingDB in 2015: A
public database for medicinal chemistry, computational chemistry and systems pharmacology.
Nucleic Acids Res 44:D1045–D1063
22. Chen X, Liu M, Gilson MK (2001) Binding DB: a web-accessible molecular recognition
database. Combi Chem High-Throughput Screen 4:719–725
23. de Matos P, Alcántara R, Dekker A, et al (2010) Chemical entities of biological interest: an
update. Nucleic Acids Res 38(Database):D249–D254
24. Mendez D, Gaulton A, Bento AP et al (2019) ChEMBL: towards direct deposition of bioassay
data. Nucleic Acids Res 47:D930–D940
25. Davies M, Nowotka M, Papadatos G et al (2015) ChEMBL web services: streamlining access
to drug discovery data and utilities. Nucleic Acids Res 43:W612–W620
26. Chen JH, Linstead E, Swamidass SJ, Wang D, Baldi P (2007) ChemDB update-full-text search
and virtual chemical space. Bioinformatics 23:2348–2351
27. Pence HE, Williams A (2010) ChemSpider: an online chemical information resource. J Chem
Educ 87:1123–1124
28. Feng Z, Chen L, Maddula H, Akcan O, Oughtred R et al (2004) Depot: a data warehouse for
ligands bound to macromolecules. Bioinformatics 20(13):2153–2155
29. National Cancer Institute, Washington, DC (1997). http://rex.nci.nih.gov. Accessed on 15 Oct
2022
30. Kaiser J (2005) Science resources. Chemists want NIH to curtail database. Science
308(5723):774
31. Michalsky E, Dunkel M, Goede A, Preissner R (2005) SuperLigands—A database of ligand
structures derived from the Protein Data Bank. BMC Bioinformatics 6:122
32. Wishart D, Arndt D, Pon A, (2015) T3DB: the toxic exposome database. Nucleic Acids Res
43(Database issue):D928–D934.
33. Sterling T, Irwin JI (2015) ZINC 15—Ligand discovery for everyone. J Chem Inf Model
55:2324–2337
34. Singla D, Sharma A, Kaur J, Panwar B, Raghava GP (2010) BIAdb: a curated database of
benzylisoquinoline alkaloids. BMC Pharmacol 10:4
35. Dictionary of natural products online. http://dnp.chemnetbase.com. Accessed on 15 Oct 2022
36. Mangal M, Sagar P, Singh H, Raghava GP, Agarwal SM (2013) NPACT: naturally
occurring plant based anti-cancer compound-activity-target database. Nucleic Acids Res
41(Database):D1124–D1129
37. McDonagh EM, Whirl-Carrillo M, Garten Y, Altman RB, Klein TE (2011) From phar-
macogenomic knowledge acquisition to clinical applications: the PharmGKB as a clinical
pharmacogenomic biomarker resource. Biomark Med 5(6):795–806
38. Yan D, Zheng G, Wang C et al (2020) HIT 2.0: an enhanced platform for Herbal Ingredients’
targets. Nucleic Acids Res 50:D1238–D1243
39. Madej T, Addess KJ, Fong JH et al (2012) MMDB: 3D structures and macromolecular
interactions. Nucleic Acids Res 40(Database):D461–D464
40. Berman H, Henrick K, Nakamura H, Markley JL (2007) The worldwide Protein Data
Bank (wwPDB): ensuring a single, uniform archive of PDB data. Nucleic Acids Res
35(Database):D301–D303.
41. Zhou Y, Zhang YT, Lian XC et al (2022) Therapeutic target database update 2022: facili-
tating drug discovery with enriched comparative data of targeted agents. Nucleic Acids Res
50(D1):1398–1407
298 S. Kar and J. Leszczynski

42. The, UniProt, Consortium (2021) UniProt: the universal protein knowledgebase in 2021.
Nucleic Acids Res 49:D480–D489
43. Wang G, Li X, Wang Z (2016) APD3: the antimicrobial peptide database as a tool for research
and education. Nucleic Acids Res 44:D1087–D1093
44. Waghu FH, Idicula-Thomas S (2020) Collection of antimicrobial peptides database and its
derivatives: Applications and beyond. Protein Sci 29(1):36–42
45. Tyagi A, Tuknait A, Anand P et al (2015) CancerPPD: a database of anticancer peptides and
proteins. Nucleic Acids Res 43(Database issue):D837–D843
46. Wang J, Yin T, Xiao X, He D, Xue Z, Jiang X, Wang Y (2018) StraPep: a structure database
of bioactive peptides. Database 2018:bay038
47. Schellenberger J, Park JO, Conrad TM et al (2010) BiGG: a biochemical genetic and genomic
knowledgebase of large scale metabolic reconstructions. BMC Bioinf 11:213
48. King ZA, Lu JS, Dräger A et al (2016) BiGG models: a platform for integrating, standardizing,
and sharing genome-scale models. Nucleic Acids Res 44(D1):D515–D522
49. Karp PD, Billington R, Caspi R et al (2019) The BioCyc collection of microbial genomes and
metabolic pathways. Brief Bioinform 20:1085–1093
50. Wishart DS, Guo AC, Oler E, et al. (2022) HMDB 5.0: the Human Metabolome Database for
2022. Nucleic Acids Res 50(D1):D622–D631
51. Kanehisa M, Goto S (2000) KEGG: Kyoto encyclopedia of genes and genomes. Nucleic Acids
Res 28:27–30
52. Kenneth Haug K, Keeva Cochrane K, Venkata Chandrasekhar Nainala VC et al (2020) Metabo-
Lights: a resource evolving in response to the needs of its scientific community. Nucleic Acids
Res 48(D1):D440–D444
53. Kale NS, Haug K, Conesa P et al. (2016) MetaboLights: an open-access database repository
for metabolomics data. Curr Protoc Bioinf 53:14.13.1–14.13.18
54. Wishart DS, Frolkis A, Knox C et al. (2010) SMPDB: the small molecule pathway database.
Nucleic Acids Res 38(Database issue):D480–D487
55. Jewison T, Su Y, Disfany FM, et al. (2014) SMPDB 2.0: Big improvements to the small molecule
pathway database. Nucleic Acids Res 42(Database issue):D478–D484
56. Kutmon M, Riutta A, Nunes N et al (2016) WikiPathways: capturing the full diversity of
pathway knowledge. Nucleic Acids Res 44(D1):D488–D494
Index

A Cerebrospinal Fluid (CSF), 138

Absorption, Distribution, Metablolism, Clinical Proteomic Tumor Analysis
Excretion, and Toxicity (ADMET), Consortium (CPTAC), 277
4 Collection of Anti-Microbial Peptides
Acetylcholine (ACh), 57 (CAMPR3), 286
Acetylcholinesterase inhibitors (AChEIs), Computer-Aided Drug Design (CADD), 29
57 Computer-Aided Drug Designing (CADD),
Acquired immune deficiency syndrome 270
(AIDS), 158 Constraint Based Reconstruction and
Adverse Outcome Pathways (AOPs), 240 Analysis (COBRA), 289
African Green Monkeys (AGM), 143
Alzheimer’s disease (AD), 57
Aminotransferase (ALT), 199 D
Amyloid precursor protein (APP), 58 Dictionary of Organic Compounds (DOC),
Antimicrobial Peptide Database (APD), 280
286 Dynein Motor Binding region (DMB), 29
Applicability Domain Index (ADI), 218
Artificial Intelligence/Machine Learning
(AI/ML), 9 E
Artificial Neural Network (ANN), 127, 183 Electron Density scores for Individual
Aspartate aminotransferase (AST), 199 Atoms (EDIA), 10
Encephalo Myocarditis Virus (EMC), 124
European Chemical Agency (ECHA), 226
B European Food Safety Authority (EFSA),
Biochemical Genetic and Genomic (BiGG), 221
289
Biosafety Level-4 (BSL-4), 138
F
Free Energy Perturbation (FEP), 16, 169
C
Cancer Data Access System (CDAS), 277
Carcinogenic, Mutagenic, or Reprotoxic G
(CMR), 223 Gaussian accelerated Molecular Dynamics
Catalytic Anionic Site (CAS), 63 (GaMD), 169
Central Nervous System (CNS), 58 Genomic Data Commons (GDC), 277

© The Editor(s) (if applicable) and The Author(s), under exclusive license 299
to Springer Nature Switzerland AG 2023
S. Kar and J. Leszczynski (eds.), Current Trends in Computational Modeling for Drug
Discovery, Challenges and Advances in Computational Chemistry and Physics 35,
https://doi.org/10.1007/978-3-031-33871-7
300 Index

Glycogen synthase kinase 3 beta Mild Cognitive Impairment (MCI), 58

(GSK-3β), 62 Molecular Dynamics (MD), 17, 120
Granulocyte Colony-Stimulating Factors Molecular Mechanics (MM), 118
(G-CSF), 197 Molecular Mechanics Poisson-Boltzmann
Graphical Processing Units (GPUs), 17 Surface Area (MM-PBSA), 151, 169
Grid Independent Descriptors–GRIND, 43 Multiple Linear Regression (MLR), 118,
Ground-Glass Opacity (GGO), 199 179
Multi-Target Drugs (MTDs), 60

H
Heat shock protein (Hsp90), 28 N
Heat shock transcription facto-1 (HSF-1), Naive Bayes (NB), 118
28 National Institutes of Health (NIH), 276
Hendra Virus (HeV), 137 Naturally Occurring Plant-Based
Hepatitis C virus (HCV), 142 Anti-Cancer
Highly active antiretroviral therapy Compound-Activity-Target
(HAART), 157 Database (NPACT), 280
High Throughput Screening (HTS), 2, 140 Neuraminidase (NA), 125
Histone acetyltransferases (HATs), 26 Neurofibrillary Tangles (NFTs), 58, 59
Histone deacetylase 10 (HDAC10), 28 New Chemical Entities (NCEs), 3
Histone deacetylases (HDACs), 26 Nipah Virus (NiV), 137
Human Immunodeficiency Virus-1 N-methyl-D-aspartate (NMDA), 58
(HIV-1), 157 N-methyl-D-aspartate receptor (NMDAR),
Human Immunodeficiency Virus (HIV), 57
111 Non-Nucleoside Reverse Transcriptase
Human Intestinal Absorption (HIA), 151 Inhibitors (NNRTI), 160
Humanized Monoclonal antibodies Nuclear Magnetic Resonance (NMR), 4
(hMAbs), 145 Nucleoside or nucleotide reverse
Human Metabolome Database (HMDB), transcription inhibitors (NRTIs), 160
289

O
I Organization for Economic Co-operation
International Committee on Taxonomy of and Development (OECD), 240
Viruses (ICTV), 196
International Conference on Harmonisation
(ICH), 215 P
Pathway/Genome Databases (PGDBs), 289
Pearson’s Correlation Coefficient (PCC),
K 148
k-Nearest Neighbor (kNN), 36, 118 Persistent, Bioaccumulative and Toxic
Kyoto Encyclopedia of Genes and (PBT), 223
Genomes (KEGG), 290 Perturbation Theory and Machine Learning
(PTML), 129
Pharmacogenomics Knowledgebase
L (PharmGKB), 282
Ligand-Based Drug Design (LBDD), 116, Physiologically-Based Pharmacokinetic
269 (PBPK), 240
Presenilin 1 (PSEN1), 58
Principal Component Analysis (PCA), 118
M Protein Data Bank (PDB), 4, 119, 284
Machine Learning (ML), 37 Proteolysis Targeting Chimera (PROTAC),
Macromolecular X-ray crystallography 121
(MX), 4 Pseudotyped Virus (pVSV), 140
Index 301

Q S
Quantitative predictions using the RASAR Severe Fever with Thrombocytopenia
approach (q-RASAR), 239 Syndrome (SFTS), 196
Quantitative Reverse Severe Fever with Thrombocytopenia
Transcription-Polymerase Chain Syndrome Virus (SFTSV), 196
Reaction (qRT-PCR), 144 Simplified Molecular-Input Line-Entry
Quantitative Structure Activity Relationship Systems (SMILES), 147
(QSAR), 29, 147, 231, 273 Single-stranded RNA (ssRNA), 138
Small Molecule Pathway Database
Quantum Mechanics (QM), 118
(SMPDB), 292
Quantum Mechanics/Molecular Mechanics
Structural Alerts (SA), 215
(QM/MM), 16
Structure Activity Relationships (SAR), 4
Quantum Polarized Ligand Docking Structure-Based Drug Design (SBDD), 2,
(QPLD), 171 116, 269
Structure database of Bioactive Peptides
(StraPep), 288
Support vector machines (SVM), 36
R
Radius of gyration (Rg), 45 T
Random Forest (RF), 119 Therapeutic Target Database (TTD), 284
RApid DEcoy Retriever (RADER), 148 Three-dimensional (3D), 2
Rational Drug Design (RDD), 29 Toxin and Toxin-Target Database (T3DB),
Reactive Oxygen Species (ROS), 59 276
Read-Across Structure-Activity Tumour Necrosis Factor (TNF), 197
Relationship (RASAR), 239
Real Space Correlation Coefficient
U
(RSCC), 10
Universal Protein resource (UniProt), 285
Registration, Evaluation, Authorization,
US Environmental Protection Agency (US
and restriction of Chemicals
EPA), 240
(REACH), 240
Regulated on activation and generally
expressed by T-cells (RANTES), V
198 Virtual Screening (VS), 36
Respiratory Syncytial Virus (RSV), 115
Ribonuclease Targeting Chimera
(RIBOTAC), 121 W
Root Mean-Squared Deviation (RMSD), 45 World Drug Index (WDI), 36