Ask a general AI model how many approved drugs hit a target, and it might tell you three when the real answer is six, sounding just as confident either way.

If your team is grounding drug discovery decisions in AI output with no way to trace where the answer came from, you're one regulator's question away from a very expensive problem.

Lisa Downey is CEO of DrugBank, a structured biomedical intelligence platform cited in more than 60,000 papers and used by nine of the top 20 global pharma companies. She previously built Clarivate's genomic and rare disease data business from the ground up and held leadership roles at GlobalData, giving her almost 20 years across healthcare and life sciences data.

Lisa breaks down why the bottleneck in AI-driven drug discovery has shifted from data scarcity to trustworthy grounding, and what that means for teams making target identification and go/no-go calls. You'll hear how DrugBank's knowledge graph separates causation from correlation, why reproducibility matters more than speed, and what questions to ask before building a reference data layer in-house.

This episode covers deterministic versus probabilistic data, human-in-the-loop versus human-over-the-loop curation, and how biopharma teams connect grounding layers to their AI agents through MCP. It's built for data and analytics leaders, R&D teams, and anyone deciding whether to build or buy their biomedical data infrastructure.

Key Takeaways

- A general model asked how many approved drugs hit PD-L1 will answer with total confidence, and total inaccuracy, missing half the real number without any signal that it's wrong.

- Anthropic's own benchmarks found frontier models pulling public genomic data got it right as little as 17% of the time, until a deterministic tool pushed accuracy past 99%.

- DrugBank moved from human-in-the-loop curation to human-over-the-loop oversight once its data was connected enough that one expert validating one relationship could cascade trust across dozens of related facts.

- Before building or buying a reference data layer, Lisa lays out four questions that separate real infrastructure from marketing, starting with whether every fact traces back to a source and a date.

Chapter Markers

00:00 Why data scarcity isn't the real bottleneck anymore

01:22 What drew Lisa to DrugBank's mission

03:04 What DrugBank is and who relies on it

05:03 The grounding layer: completeness and reproducibility

07:28 Anthropic's benchmark on data infrastructure

09:21 The high-stakes decisions DrugBank data informs

12:23 Where lost cycle time actually comes from

14:13 DrugBank versus homegrown knowledge graphs

19:43 Human-in-the-loop versus human-over-the-loop curation

24:12 How DrugBank checks its own data quality

25:37 Deterministic versus probabilistic data explained

28:52 The J&J case: separating causation from correlation

33:06 Connecting DrugBank to your AI stack via MCP

37:28 Four questions to ask before you build or buy

42:15 Where DrugBank fits, and where it doesn't

44:44 AI as an amplifier of both good and bad decisions

Useful Links & Resources

- Connect with Lisa Downey on LinkedIn (https://www.linkedin.com/in/lisaldowney/)

Connect With the Show

- Ross Katz on LinkedIn (https://www.linkedin.com/in/b-ross-katz/)

- CorrDyn on LinkedIn (https://www.linkedin.com/company/corrdyn/)

Have you run into an AI model giving you a confident, wrong answer in your own R&D work? Tell us about it in the comments, we're always looking for real examples for future episodes.

Visit corrdyn.com to learn how CorrDyn can help your organization extract value from data.

#DataInBiotech #BiotechAI #DrugDiscovery #DataScience #LifeSciences

Most drug discovery genomic data comes from a thin slice of the world, and that bias follows every decision downstream.

Your team can run a Mendelian randomization study on 35,000 patients and still walk away with a single signal that doesn't even apply to the population you care about. If your phenotype definitions are fuzzy, more data won't save you.

Erika Kvikstad is a computational biologist who led precision medicine for cardiovascular disease at Bristol-Myers Squibb, working on therapies including Camzyos for hypertrophic cardiomyopathy. She now works independently on genomic data equity, focused on how reference populations shape everything from target discovery to clinical trial recruitment.

You'll get a practical look at how to evaluate real-world data vendors, why heart failure is nearly impossible to define cleanly from billing codes, and where statistical power breaks down even with tens of thousands of patients. Erika also explains how her team used AI to reconstruct missing imaging data and validate cardiomyopathy diagnoses at scale.

This episode covers GWAS studies, Mendelian randomization, UK Biobank, proteome-wide analysis, and the practical gap between biobank-scale data and disease-specific cohorts. It's built for data and analytics leaders working in life sciences who need to understand where genomic bias enters their pipeline, not just that it exists.

Clarification

Around 57:58–58:24, in discussing the proteome-wide Mendelian randomization study, Erika moved quickly between two related findings. BTN3A2 was identified as a candidate associated with ischemic stroke and potential immune-modulatory biology. Separately, single-cell expression data helped contextualize other candidate signals, including some with enriched expression in cardiomyocyte populations. Cardiomyocyte-enriched expression was not a specific finding for BTN3A2.

Chapter Markers

00:00 Whose genome are we designing drugs for

01:34 Erika's path from academic genomics to BMS

03:48 Building the precision medicine strategy at BMS

06:37 Ross shares his own HCM diagnosis

07:09 Why heart failure resists clean definition

11:11 How medication use reclassifies patients

14:35 Imaging as a biomarker, and its data gaps

20:23 Data infrastructure gaps across regions

22:44 What to look for when evaluating a data vendor

27:35 Consortia and biobanked specimens for rare mutations

29:52 Cardiovascular data infrastructure versus oncology

32:29 Where statistical power breaks down

37:07 UK Biobank's strengths and its limits

40:01 Bridging broad biobanks with disease-specific cohorts

44:32 How reference population bias propagates downstream

48:53 Where genomic bias hits hardest in the pipeline

53:18 Inside a proteome-wide Mendelian randomization study

59:42 Choosing the right computational tool for the question

1:06:38 Building globally representative genomic infrastructure

1:08:04 Ross's takeaways on bias and statistical power

Useful Links & Resources

- Erika on LinkedIn: https://www.linkedin.com/in/erikakvikstad

- UK Biobank: https://www.ukbiobank.ac.uk

- Alliance for Genomic Discovery: https://alliancegenomicdiscovery.org

- SHaRe Registry (DCM Foundation): https://dcmfoundation.org

Connect With the Show

- Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/

- (Ross Katz on X: https://x.com/brosskatz

- CorrDyn LinkedIn: https://www.linkedin.com/company/corrdyn/

Have you run into genomic reference bias in your own work? Tell us what it looked like and how your team caught it.

Visit corrdyn.com to learn how CorrDyn can help your organization extract value from data.

Subscribe to Data in Biotech so you don't miss the next conversation.

Why treating the cell, not the protein, could turn chronic disease treatment into something closer to a cure.

You've built single-cell pipelines that spit out clusters, p-values and target lists, but nothing that survives contact with the clinic. What if the clustering method itself is quietly leading you astray?

Adam Freund is Founder and CEO of Arda Therapeutics, a biotech using single-cell sequencing to find the pathogenic cells driving chronic disease. He spent seven years as a Principal Investigator at Calico Life Sciences, building a research lab on the biology of ageing and helping grow the company from 15 to more than 200 people, and holds a PhD in Molecular and Cell Biology from UC Berkeley.

You'll get a working model for how Arda's discovery engine turns single-cell and spatial transcriptomic data into causal cell targets. Adam explains why a common statistical shortcut in single-cell analysis produces disease signals that don't hold up and how cell depletion could replace daily dosing with a handful of treatments that reset the immune system.

Ross and Adam cover how Arda finds pathogenic cell populations across hundreds of donors, why chi-squared tests on cell clusters can substitute cell count for donor count without anyone noticing, and how B-cell depletion therapies proved that removing a cell can beat blocking its pathway. This one is for data science leaders and computational biologists building single-cell pipelines, not listeners after a general intro to drug discovery.

Key Takeaways

- Chi-squared tests on cell clusters draw their statistical power from the number of cells, not the number of donors, so a single oversampled patient can produce the same p-value as a hundred-donor study.

- Rituximab clears 100% of B cells from circulation yet does nothing for lupus because the disease-driving cells live in tissue, not blood, a lesson now shaping where Arda tests its own molecules.

- Neighborhood analysis scores each cell by the donor identity of its nearest neighbours rather than forcing cells into predefined clusters, producing a continuous disease-enrichment map with no cluster boundaries.

- When depleted cells regrow, they often come back without the trait that made them harmful in the first place, which means a handful of doses can hold a chronic disease in remission for months.

Chapter Markers

00:00 Why cell depletion beats pathway blocking

01:05 Welcome Adam Freund to the show

01:30 From Calico Life Sciences to founding Arda

03:29 Why blocking one pathway rarely works

05:32 B-cell depletion as the proof of concept

08:14 Building a modular library of depletion tools

10:46 Single-cell sequencing removes the need for a hypothesis

11:43 Why clustering is a dial, not ground truth

15:24 The chi-squared trap in single-cell analysis

20:40 Neighbourhood analysis and donor-weighted scoring

23:44 Moving from enrichment to causality

26:32 Inside Arda's lead fibrosis program

30:33 Why solid tissue testing beats blood samples

34:25 Simulating depletion in spatial transcriptomic data

38:49 The case for intermittent dosing over daily pills

43:58 The data infrastructure behind Arda's platform

48:46 Where spatial and protein data are heading

Useful Links & Resources

- Adam Freund on LinkedIn: https://www.linkedin.com/in/adam-freund-0657654

- CorrDyn: https://corrdyn.com

Connect With the Show

- Host Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/

- Host Ross Katz on X: https://x.com/brosskatz

- CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/

If your team runs single-cell pipelines, how do you currently decide on the number of clusters, and have you ever checked whether your significance scales with donor count rather than cell count? Tell us in the comments; we're building a running list of data QA checks for biotech data science teams.

Visit corrdyn.com to learn how CorrDyn can help your organization extract value from data.