Genesis Molecular AI co-founder Evan Feinberg and CTO Sergey Edunov (former Meta Llama lead) discuss how their PEARL model is advancing protein-ligand co-folding for small molecule drug discovery. Key topics include why 3D structure prediction is the most innovative area of diffusion research today, why the standard 2Ã… RMSD benchmark is misleading and 1Ã… is the real threshold for useful predictions, and how PEARL achieved zero-shot top performance on the OpenBind benchmark by accurately modeling induced fit without MD simulations. They also discuss SAPPHIRE, their internal agentic drug discovery system that can now iterate like a medicinal chemist, and why tight integration of ML, large-scale compute, and real-world drug discovery data is the key enabler that academia struggles to replicate.
