Episode
Synthesizable by Design: Rethinking AI's Role in Small Molecule Drug Discovery
- Podcast
- Data in Biotech
- Published
- Jun 17, 2026
- Duration seconds
- 3569
- Processing state
not_requested- Canonical source
- https://www.corrdyn.com/
Actions
POST https://stenobird.com/v1/public/podcasts/data-in-biotech-6628050/episodes/synthesizable-by-design-rethinking-ai-s-role-in-small-molecule-drug-discovery/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/data-in-biotech-6628050/synthesizable-by-design-rethinking-ai-s-role-in-small-molecule-drug-discovery.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
In this episode of Data in Biotech, host Ross Katz sits down with Paul Finn, Chief Scientific Officer at Oxford Drug Design, for a conversation on what it actually takes to find a drug molecule that works not just on paper but also in the lab, in the cell, and, ultimately, in the clinic. Paul brings four decades of experience across what became GSK, Pfizer, and a series of Oxford-area spinouts and has shepherded a compound all the way to a marketed drug. That perspective gives him a particular kind of skepticism toward AI results that look too good to be true because he's done the work of checking whether they are. The conversation moves through synthesizability as a first-class constraint, why chemistry has proven so much harder for AI than biology, how 3D molecular representation gets closer to the physics that actually matters, and what rigorous multi-parameter optimization looks like when you're trying to kill cancer cells and drug-resistant bacteria at the same time. What you'll learn in this episode: >> Why synthesizability is chronically underestimated and why changing a single atom in a structure can take a molecule from trivially easy to make to practically impossible >> How Oxford Drug Design constrains the generative search to reaction schemes and purchasable building blocks, and why that chemical space is still so vast that novelty is not meaningfully sacrificed >> Why most generative AI models learn from a 2D string representation of a molecule; two steps removed from the 3D physics that govern how a drug actually binds to its target >> How Bayesian optimization over reagent space, rather than molecular space, allows an active learning loop to focus on the structural patterns associated with activity >> Why benchmarking comple…