Where AI actually delivers in drug discovery with Dave Hallett

Recursion [RXRX] CSO Dave Hallett joined Patrick Short, CEO of Sano Genetics and host of The Genetics Podcast for a conversation on the real gains and remaining hurdles for AI drug discovery and development. Drawing on his nearly 30 year experience, beginning at Merck in the mid-1990s, Dave shared how AI has fundamentally changed what’s possible  – from research capabilities, to molecule design, to clinical trial planning – but noted that it still can’t change the laws of physics.

Highlights from their conversation include:

How AI has allowed for true multi-parameter optimization

Dave said the shift from single-parameter to true multi-parameter optimization is the biggest change he's seen at the bench. A person can hold roughly six or seven variables in their head at once, he said; a computational system can track far more, without recency bias, across active learning loops. The result: "AI has massively compressed the time and the cost that's required to get from an idea to a candidate." Recursion has seen that show up as roughly 80% reductions in compounds made and 50-70% cuts to time-to-candidate on affected programs.

Simulating trials before running them

Rather than copying existing clinical trial protocols, Dave shared how Recursion pressure tests inclusion and exclusion criteria through simulation before a study launches — asking which criteria are actually still fit for purpose. Applied to REC-7735, evaluated in the ZINNIA trial, that process expanded the pool of patients the trial could reach by approximately 20%. 

Looking forward with pragmatism and optimism

Dave was candid about AI’s limitations, too, noting that the industry is "still a long way" from modeling emergent human pathophysiology at the organism scale. He also discussed Recursion's partnership work on the breakthrough neuro map, and offered advice to early-career scientists: build deep domain expertise while remaining "bilingual" across biology, chemistry, and computation.