Lindsay Edwards went from producing UK top ten singles with Sting and Whitney Houston to pioneering the first data science group at GSK — and he's carried the same creative instinct into machine learning ever since. Now CTO of Relation Therapeutics, he's one of the clearest-eyed voices in the industry on where AI in biology is genuinely working and where it's still mostly noise.

Lindsay joined Benchling CEO Sajith Wickramasekara to talk about why "virtual cells" are dangerously overpromised, why legacy data is usually worth less than pharma companies think, and what it will actually take for ML to earn its place in drug discovery.

Key Takeaways:

➜ "Virtual cell" has become a wildly overpromised term — what most researchers actually mean is predicting gene expression, which remains largely unsolved and is nowhere close to modeling the full complexity of a cell.

➜ Biology's data bottleneck isn't just a volume problem: because measurements are made in arbitrary relative units rather than absolute values, models routinely learn batch effects instead of meaningful biology.

➜ Rather than spending millions curating legacy data that may be fundamentally unmeasurable, companies should invest that same money in ensuring every new clinical study is collected consistently — and within three years, they'd have an industry-defining dataset.

Chapters:

[00:51] From chart-topping musician to machine learning pioneer [12:31] Why AI needs a reality check [19:54] The biggest gaps in AI today [22:46] Why biologists need to be more like physicists [26:05] Why too much compute is a problem [32:19] Why legacy data is worth less than you think [34:52] The future of machine learning in biology

About Lindsay:

Lindsay Edwards is Relation’s CTO and Head of Platform. Previously, Lindsay was VP and Head of AI for Respiratory and Immunology at AstraZeneca. Originally a specialist in systems biology, he joined GlaxoSmithKline in 2014 from the Physiology faculty at King’s College London. He started GSK’s first Data Science group, was Head of Respiratory Data Sciences, Global Head of Respiratory Digital, Data, and Analytics, and then VP and Head of AI/ML for the UK and Europe before becoming VP of AI/ML Engineering. He holds a DPhil in Physiology from Oxford, is a scientific advisor to the Eric and Wendy Schmidt Centre at MIT, and has published more than 40 articles in peer-reviewed journals.

💡 Learn more about Relation Therapeutics: www.relationrx.com

Guest Highlights:

"My main objection is when someone says, 'we're using AI for drugdiscovery.' That's like saying, 'I'm using chemistry for drug discovery.'"

"When somebody says 'virtual cell,' I imagine laypeople think of a computational model that synthetically captures all of the complexity of a cell. We are light years away from that."

"Rather than spending $2 million trying to figure out how to drag marginal value out of old data, just spend that money making sure that every single clinical study you run from now is consistent. Within three years, you'd have an industry-defining dataset."

🔗 Links:

➜ Connect with Sajith: linkedin.com/in/sajithw ➜ Connect with Lindsay: linkedin.com/in/lindsay-edwards-7a7268a7