RECEPTOR.AI

Ever wanted to tell ChatGPT: “Pretend you’re a molecular dynamics engine, run my target and give me representative conformations”? A team at the University of Cambridge just took a step in that direction. They featured MD-LLM-1: an LLM trained on short MD simulations. It discretizes each protein frame into tokens, predicts the next-frame tokens, and decodes them into a 3D protein conformation From a single target state it can propose alternative, physically plausible conformations. There is only early proof of concept but it looks interesting and shows an unconventional direction for using LLM architectures. At Receptor.AI, we track the shift from physics-based simulations to AI-driven methods. After our recent benchmark of conformational sampling methods, AlphaFold Sample 2 and BioEmu, fell short of expectations (link in the comments), we’re watching for new approaches like this one. #drugdiscovery #llm #moleculardynamics

Pranam Chatterjee and colleagues recently introduced PepMLM, a model that designs peptide binders directly from protein sequences, showing how sequence-based AI can infer binding patterns without structural input. The work highlights both the promise and the limitation of sequence-only approaches. Binding does not equal therapeutic effect: stability, bioavailability, and binding mode specificity are the real bottlenecks. This is where Receptor.AI comes in. Our peptide platform includes multi-parameter AI optimization that enables us not only to generate binders, but also to predict and improve their activity and ADMET/PK parameters. Read more about our peptide approach: https://lnkd.in/eemzfHrx #drugdiscovery #peptide #ai

Receptor.AI has entered a collaboration with Oragenics, Inc, a US-based clinical-stage biopharmaceutical company developing treatments for neurological disorders. As part of the project, Receptor.AI will apply its AI platform to analyze Oragenics’s molecular portfolio and generate hypotheses on receptor binding relevant to conditions such as Alzheimer’s disease, dementia, PTSD, and anxiety. In this collaboration, Receptor.AI will apply its integrated workflow combining LLM-based target profiling, proteome-wide drug–target interaction screening, and ArtiDock AI docking with comprehensive PK prediction using the ADMETiQ model family. The collaboration aims to improve the efficiency of early-stage discovery by linking AI predictions with laboratory validation in a systematic workflow across CNS-related indications, enabling more efficient identification of therapeutic candidates for neurological disorders. #ai #drugdiscovery #neurology #pharma

Novel AR antagonist targets the dimerization interface, bypassing resistance mutations A recent study reports N-(thiazol-2-yl) furanamide derivatives as potent oral androgen receptor (AR) antagonists. Most AR antagonists work at the ligand-binding pocket and often cross the blood–brain barrier, causing CNS side effects such as seizures. The lead compound, C13, is different on two counts: it binds the AR dimerization interface pocket, bypassing common resistance mutations, and it was engineered for low brain penetration, pointing toward a peripherally selective AR therapy. At Receptor.AI, we see the same principle at play. Drug design isn’t only about where a compound binds, but also where it distributes. Our ADMET model family generates predictions that guide how we optimize chemical structures, helping us reduce the likelihood of exposure in unwanted compartments, whether lowering BBB permeability or avoiding metabolic pathways linked to toxicity. #drugdiscovery #prostatecancer #biotech

High-quality experimental data, especially protein–ligand complex structures, is bedrock for successful drug discovery. That’s why Cryo-EM density maps, 3D maps of electron-scattering potential that reveal where atoms presumably are located, become essential for structure-based drug design. But while they deliver near-atomic resolution for proteins, bound ligands frequently remain under-resolved, which complicates their interpretation. A team at Stockholm University published a practical end-to-end pipeline for refining ligand poses: they predicted the complex with Chai-1, fitted that model into the Cryo-EM density map, and ran density-guided molecular dynamics (MD biased by the density map so atoms are pulled into observed density without breaking stereochemistry and connectivity). Across 10 protein–ligand complexes, AI-predicted structures correlated with density maps at 40-71%. After MD refinement of these structures, the correlation increased to 82-95%. At Receptor.AI, we perform density-guided molecular dynamics using experimental density maps obtained from X-ray crystallography or other methods as well. It helps us get reliable ligand binding poses for binding mode analysis and further virtual screening. #drugdiscovery #CryoEM #biotech

Does machine learning–based protein conformational sampling actually improve protein–protein docking? The answer: not really. In our recent study (link in comments), we tested AFSample2 and BioEmu on 30 complexes from the PINDER-AF2 Apo benchmark, generating more than 43,000 docked poses. Here's what we learned: ▪️ Most sampled conformations were no closer to the bound state than Apo (unbound state). ▪️ Even when near-native poses appeared, current scoring functions failed to bring them to the top.  ▪️ Sometimes, more sampling just meant more noise. Both steps hold us back: producing near-native conformations and identifying them among thousands of predictions. 👉 If you’re interested in where rigid docking actually breaks down and what directions could move the field forward, this study may be worth a look. #PPI #drugdiscovery #docking

Our CSO, Dr. Semen Yesylevskyy, attended the 10th Lipidomics Forum in Vienna (14–17 September). Modern lipidomics is becoming a valuable tool in drug discovery, and the growing scale and complexity of these datasets make them well-suited for machine learning. Dr. Yesylevskyy presented work showing that approved amphiphilic drugs can exhibit unexpected membranotropic behavior that may affect their distribution, receptor binding, off-target activity, and ADMET properties. At Receptor.AI, we see strong potential where lipid environments influence drug action, from membrane proteins to membrane-targeting compounds. Accounting for these interactions opens new space for understanding therapeutic effects and safety. #lipidomics #drugdiscovery #ai

Receptor.AI is looking for an accomplished scientific leader to join our executive team as Head of Drug Discovery. The position will play a central role in advancing our discovery projects, guiding multidisciplinary teams, and driving scientific and operational excellence across the organization. If you have deep experience in drug discovery and a track record of advancing molecules toward the clinic, we’d love to hear from you. #drugdiscovery #biotech #hiring

Excited to share that I’ll be in Boston on September 16–17, 2025 for the LSX World Congress USA. I’ll also attend the Obesity Science & Innovation, where I’ll be discussing Receptor.AI’s expertise in obesity targets, particularly GPCRs and GLP-1, across peptide and small molecule drug modalities. If you’ll be at either event, I’d be glad to connect and talk about recent developments in drug discovery. More info: https://lnkd.in/ea3r8JwS #LSXUSA #DrugDiscovery #AI #Biotech #Obesity

Hydrophobic pockets often hide a crucial player: water. In human cannabinoid receptor 2 (CB2R), a recent study shows that placing predicted water molecules in the pocket reveals hydrogen-bond bridges and chemotypes that “dry” screens miss. From a 10M library, the team confirmed selective CB2R ligands with the negative logarithm of the inhibitory constant (pKi) up to 7,70. Water-aware docking isn’t new, but here it enabled targeting a difficult hydrophobic pocket and surfaced a novel scaffold. At Receptor.AI, we also customize pockets with water molecules when needed, but this step erodes the accuracy of classical docking. Our AI technology for protein–ligand docking, ArtiDock, closes this gap. Check the link in comments for a benchmark. #drugdiscovery #docking #biotech