From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio

From a four-billion-measurement Kaggle competition to models that recognize a molecule’s chemist, Leash Bio is testing what AI actually learns about chemistry. Co-founder and CEO Ian Quigley joins Brandon Anderson and RJ Honicky to explain how Leash makes DNA-encoded libraries (DELs) produce clean, reproducible protein–small-molecule binding data—and why collecting the right experiments matters as much as choosing a model. We go deep on hidden shortcuts in public datasets, Hermes’ comparison with Boltz, and the protein mutations that expose memorization. Ian explains how Leash’s measured chemical landscapes can become a sandbox for reasoning agents, with rapid experimental feedback and thousands of medicinal chemistry campaigns. We also examine the limits of DELs, moving hits toward medicines, the economics of AI drug discovery, and the experimental bottlenecks that still need solving. We discuss: • How DNA-encoded libraries use barcodes and sequencing to screen millions of molecules • Catching DNA-binding artifacts with multiple barcodes per molecule • Why Leash prioritizes reproducible measurements across many protein targets • The BELKA Kaggle competition: four billion measurements, around 2,000 teams, and no better than chance on an unseen chemical library • Clever Hans in ChEMBL: learning who made a molecule instead of the chemistry • Hermes, simple baselines, careful data splits, and comparisons with Boltz • Activity cliffs, engineered protein mutations, and models that memorize protein “nicknames” • Agents reasoning about structure–activity relationships in plain English • More than 10,000 medicinal chemistry campaigns in a weekend • Going from DEL hits to mouse PK/PD, and the case for pills on difficult targets • The Kosuri paradox: if you can collect the data, why do you need the model? • Experimental know-how, detecting misfolded proteins, and scaling purification — Show Notes: https://www.latent.space/publish/post/220316507 Ian Quigley — Leash Bio • LinkedIn: https://www.linkedin.com/in/iquigley/ • Leash Bio: https://leash.bio/ • Leash blog: https://leashbio.substack.com/ Related Minos release • Blog: https://leashbio.substack.com/p/making-reality-replayable • Preprint: https://github.com/Leash-Labs/minos-paper/blob/trunk/Minos_Leash.pdf • Data and environments: https://strelka.leash.bio/ Timestamps 0:00 Hook: Learning the Chemist, Not the Chemistry 1:22 Introduction: Ian Quigley and Leash Bio 2:40 How DNA-Encoded Libraries Work 5:25 Sequencing as a Scalable Biological Readout 9:14 Barcodes, Controls, and DNA-Binding False Positives 11:51 DNA Tags, Off-DNA Validation, and Reproducibility 15:24 Why Leash’s DEL Strategy Works 19:19 The Recursion Mafia and Experimental Rigor 20:51 Why Small-Molecule Drug Discovery Matters 26:22 Screening Scale and Four Billion Measurements from a Basement 28:21 BELKA: Designing the Kaggle Competition and Hidden Evaluations 32:32 What the Kaggle Results Taught Leash 35:44 Better Data: Protein Diversity, Library Size, and Noise 40:15 Hermes: Simple Architecture and Rigorous Evaluation 46:24 Clever Hans: Predicting a Molecule’s Chemist 55:50 Testing Hermes on External Data and Comparing with Boltz 59:07 Activity Cliffs and Breaking Protein Binding Pockets 1:07:12 Reading the Mutant-Screen Results 1:11:36 Catching Hermes Memorizing Protein “Nicknames” 1:15:47 Language Models and a Scientific Reasoning Sandbox 1:21:27 Running 10,000 Medicinal Chemistry Campaigns in a Weekend 1:24:08 Chemical-Space Coverage and DEL Limitations 1:26:00 Hard Questions: Generalization and Drug-Like Chemistry 1:28:36 From DEL Hits to Mouse PK/PD in Six Months 1:32:51 Drug-Discovery Economics and Pills vs. Antibodies 1:36:32 The Kosuri Paradox: Why Build a Model If You Can Screen? 1:40:38 Chemical-Space Generalization and Predicting Potency 1:42:58 What Stops a Better-Funded Lab from Outscaling Leash? 1:44:48 Lumacaftor-Like Hits as a Misfolded-Protein Detector 1:47:01 Removing Bottlenecks: Sequencing and Parallel Purification 1:49:14 Final Takeaways and Working with Leash Sponsorship and business inquiries: business@latent.space