From algorithms to advancing care: genomics data drives progress

From algorithms to advancing care:
genomics data drives progress
Jack DiGiovanna, PhD
Festival of Genomics Boston
24 June 2025

2
Signal drives insights
Purposefully high-level as you are in diverse therapy areas, drug modalities, data types, & stages of drug discovery

Image credits: ideogram.ai 3
Create VERY LARGE datasets

Create HIGH-SIGNAL datasets

Create HIGH-SIGNAL datasets
There’s an art to designing the signal
Pathways, mechanisms?
Receptors, targets?
Data type(s)?
Analytics?
Time?

Lessons learned in data management
Confidential – for internal use only 6
From experimentalist to analyst to partner

Data is often unused because it’s not findable
Worked closely with a Top 20 pharma across multiple therapy areas, locations,
and analysis teams. Trying to incorporate population scale datasets and in-
house data to accelerate target identification.
Isn’t it expensive that one group
collects that same data when a
different group has already done this?
Yes, but it’s less
expensive than making
the data findable.

Findable, Accessible, Interoperable, & Reusable data would be great!
Significant efforts across academia and life sciences since the
publication of the FAIR guidelines (Wilkinson et al, Scientific Data, 2016)
The concepts seem simple, but it create ~15 criteria to meet.
”FAIR is a goal, not an end state”
Multiple pharma have incorporated FAIR principles into the design of
Data as a Product. This may be local or org-wide.

It’s challenging to make data FAIR… for myself
Reuse the data
& analytics?
Feb 2024
Search
usual external
hard drives
EPFL linked
Evernote and
NAS
Search
personal
Google Drive
Do my old
laptops still
boot up?
Random walk
through other
hard drives
Wait, didn't I
use Dropbox
as a post-doc?
FOUND THE
DATA &
CODE!!!

11
There is a spectrum of FAIRness and ROI is crucial
There are multiple levels of information that are searchable, also endless debates in defining "data" vs "metadata"
Dataset level Sample level
Variant or feature
level
Insight level
… dataset with AML
patients, and
both transcriptomics
and EDoH data
... samples from Male,
patients, with prior
aggressive lymphoma tr
eatment within 15
years prior to AML
diagnosis
... subjects where there is a
missense mutation in the
(ALK or KRAS), TMB >
N, environmental feature,
and non-response to first-
line therapy.
... subjects with variants also
common in NSCLC, eligible
for clinical trials within 100
miles, and likely to respond to
immunotherapy

12
Incentives are crucial to create FAIR data
NIH Data Management & Sharing Policy (2023) requires sharing but limited
• Funding or Time
• Attribution (though the s-index challenge is a step in right direction)
• Consequences
• Sustainability
Mixture of human and technology problems

13
Top-down mandates may be necessary but are not sufficient
Similarly, many large pharma have taken a top-down approach of
1. Top-down mandates across pharma that data must be shared
2. Data generators want to maintain control of their shared data and
describe their data for primary use
3. Lack of a centralized system allowing decentralized governance
4. Metadata may not be sufficient for secondary use
Mixture of human (incentives) and technology problems
“Permissions are the biggest problem today,
structure & system consistently a blocker”
Strategy & Portfolio Lead: Analytics & Insight
“Governance is always an issue and the biggest
challenge that they face”
VP Data Science
~20 FTEs in Data Governance, FAIR data and
analytics are important
Exec Dir: Data Science

14
Data integration is a consistent pain point
Detailed interviews with Top 20 (n=3) and Mid-sized
(n=3) pharma in 2025 to understand pain points with
current solutions
Data integration avg score = 9.3
Data management avg score = 8
Two of the top five pains reported for both groups
Image credit: : https://labs.openai.com/s/VtVExoasoSJuPuAl6cTTsND8
Pain adapted from: Belbury - Own work, CC BY-SA 4.0,
https://commons.wikimedia.org/w/index.php?curid=92568509

Build vs buy considerations lean towards buy
Engage with large data vendors likely have relevant data, e.g., Tempus,
Truveta, Flatiron, ConcertAI, IQVIA, Komodo, SciBrite
Data arrives assembled and often with more permissive Data Use
Agreements
Apply for access to nation-scale research datasets, e.g., UK Biobank,
FinnGen, All of Us
"The cost for my org to assemble data internally
is incalculable... I think it's infinite"
Pharma SVP at BioTechX EU 2023
(on why his org aggressively buys data)
“We want to find target engagement biomarkers
and use then guide our phase 1-2 trials. Shorten
the research time to the first human dose which
is ~1500 days and we target 500 days. It’s a
big ambition … so we are heavily investing in
cohort datasets for relevant indications."
Head of Research Solutions
(on why she needs data to achieve Corp goals)

Solving data governance, mgmt, & integration leads to great things
Calibration of a mixture-
of-experts TMB
estimator to rescue an
ICI candidate after a
failed stage with
oncology datasets
Optimization of neoantigen
workflow and epitope
prioritization leveraging
Parker Institute datasets
More effective and
efficient target
identification using
population scale
datasets
Indication expansion,
market access using claims
and EMR data
Leverage >10x more datapoints to
guide diagnosis and treatment of
aggressive pediatric cancers

With only ~200 new cases of
high-risk paediatric cancer
per year, it is imperative to
aggregate Australian data
with global data to develop
strategies to effectively treat
high-risk childhood cancer
Mark Cowley, ~2019
Image [2019] credit:
https://www.zerochildhoodcancer.org.au/about/
research---clinical-partners

19
ZERO has cutting edge somatic multi-omic workflows

20
Those portable workflows also can be brought to CBTN data

21
Bringing compute to data increased samples used for treatment by 10x
The workflows are running considerably
faster & cheaper. We have more data.
The seamless connection with NetAPP
is an absolute game-changer.
- Mark Cowley, Genomics and
Bioinformatics Lead, ZERO

22
Bringing compute to data increased samples used for treatment by 10x
PROJECT TIME

24
Key lessons learned from cases where data drove progress
FAIR is a good direction but is more complex than it seems.
Understand who your secondary users are – this can dramatically
affect metadata requirements and models.
Prospective changes are much less effort than retrospective,
spending time analyzing ROI of historical data is important.
Invest early in solid Data Governance.
Treat your data (and Analytics) as a Product.
Leveraging international standards community can be helpful for
scalability and sustainability.

25
Our mission is to
accelerate drug discovery

26
Introducing the Global Data Network (GDN)
• The GDN empowers pharma and biotech to tackle rising costs
and long timelines by leveraging fit-for-purpose data,
reducing traditional data acquisition from years to weeks.
• Previously impossible due to the lack of federated
technology to bridge the gap between providers and allow the
systemic pooling or stacking of data.
• This globally diverse provider network includes private and
public data companies/entities such as healthcare
organizations, academic research institutions, biobanks,
biospecimen providers, commercial data vendors, health
nonprofits, and national research initiatives.
• Approximately 40% of data is derived from non-US patients.
The world's largest federated data network spread across over 175M patients, 50 data providers
and hundreds of locations globally

GDN provides unique, diverse data that is curated and assembled
Data Sources +
Geographies
Data
Modalities
Therapeutic
Areas
Typical GDN datasets are
composed of longitudinal
EMR data paired with linked
additional modalities.
(e.g., Whole-exome seq,
single-cell RNA, & imaging)
Regulatory-grade,
meticulously-validated data
adheres to GDPR, HIPAA, and
global standards. Delivered in
a secure, compliant manner.
All major disease areas
covered, with a focus on
Oncology, Immunology &
Inflammation, Neurology, &
Cardiometabolic data.
Work with team of experts to
define data criteria quickly and
efficiently to expedite delivery.
Access data from 175M
patients globally.
50+ global data providers
including healthcare orgs,
biobanks, research institutions.
40% non-US sources
including Europe, Asia, Latin
America, and more.
Access to 48 Pb of public data
sets through Seven Bridges
platforms.

28
GDN provides global data for many use cases with lifetime consent
Prioritization of modalities and specific consent
mechanisms, including:
✓ Lifetime consent: Follow patients for their entire journey
✓ Direct to patient access: Recontact patients for additional data
✓ Prospective data collection: If needed
North America
South America
Asia - Pacific
Europe - Middle
East - Africa
Novel Target Identification and Disease Subtyping
Patient Cohort Identification and Trial Feasibility
Synthetic Control Arms and External Comparators
Biomarker Discovery and Validation
Real-World Evidence (RWE) for Regulatory or Label Expansion
Post-Market Safety and Effectiveness Monitoring
Digital twins
AI/ML Model Training and Validation
…

29
Our approach returns the entire dataset if available instead of
piecewise subsets

30
Our approach returns the entire dataset if available instead of
piecewise subsets
The challenges of
typical approaches
lead to 70% of
healthcare searches
going unmet

Bring us your last unmet data request
Data Sourcing
We tap into 175M+
diverse (40% non-US)
patient records, to
deliver existing
datasets or collect
bespoke cohorts.
Deliver Insights
Leverage the Velsera
Seven Bridges Platform
for secure, collaborative
analysis to derive
actionable insights for
your R&D pipeline.
Harmonization
GDN harmonizes data
subsets to meet your
needs, connecting
securely while
respecting consent and
DUAs.
Define Needs
Share your use case,
data modalities, and
cohort requirements.
We iterate with you
quickly and efficiently to
refine criteria for
optimal outcomes.
80% of requests fulfilled in a few weeks of initial inquiry into GDN
Feasibility analysis is not charged
Reduce your data acquisition time from years to weeks!

36
Find out more about Velsera & Global Data Network at booth #44
Delighted to talk data, diagnostics, or discovery
Somehow there’s also elephants…
hello@velsera.com Join our reception at Citrus & Salt tonight from 19:30-21:00

From algorithms to advancing care: genomics data drives progress

More Related Content

Similar to From algorithms to advancing care: genomics data drives progress

Recently uploaded

From algorithms to advancing care: genomics data drives progress