Physics vs. Machine Learning in Protein Design (with Actimo Labs)

Do neural networks learn the physics of proteins — or just the consequences of it? Leo Wan and Michael Holden are joined by Kailash and Jaber, co-founders of Actimo Labs (Australia), for a big-ideas conversation: why one in three antibodies is polyspecific and how that drives late-stage failure; what changed when AlphaFold 2 arrived; whether models learn physics or just its consequences; intrinsically disordered proteins and where the data runs out; the Isomorphic Labs data-vs-architecture debate; and the dark proteome we've barely begun to map. ▸ Work with RANOMICS: https://www.ranomics.com CHAPTERS 0:00 Welcome & meet Actimo Labs (Kailash & Jaber) 3:24 Antibody toxicity & polyspecificity 6:40 Physics vs. machine learning: what changed 16:54 Do models learn physics — or just its consequences? 21:51 Synthetic data & training on predictions 22:21 Isomorphic Labs: data vs. architecture 34:36 Where the next big jump comes from 35:43 World models & learning the physics 41:46 One takeaway each #proteindesign #machinelearning #AlphaFold #drugdiscovery #biotech