How Does Leash Bio Run Lab-in-the-loop Without a New Experiment?

Ian Quigley says Leash Bio runs lab-in-the-loop at compute speed by letting a language model test a binding idea against measurements the lab already collected, and get the answer back instantly.

They taught open language models medicinal chemistry in readable English, using activity cliffs: the same chemicals measured across many proteins. The model describes a structure, states the difference between two molecules, and says why the binders might be binding, including hard negatives. It then sees 10 binders and 10 non-binders that are close, writes a reason, and may pick five or 10 molecules from a catalog of millions. It looks those up in data Leash Bio already has and updates the idea. The aim is the best binder in as few steps as possible.

He calls the setup a reinforcement rollout. The small chemical changes are causal, so the paths can become reasoning traces. Agents get a budget of 100 molecules and make about a hundred calls. Plain language, he thinks, resists memorization better than a bigger Hermes, because more parameters will memorize the mutants too. A medicinal chemist might see a couple dozen of these campaigns in a career, maybe 50. Last weekend they pulled out more than 10,000 without trying hard. Inside that budget of 100, the agents reach the best binders out of 6.5 million molecules. The loop only knows chemicals already screened.