@sppatil6306@johnhanacek@NTallapragada Agreed! @kayatsenko and I got our start in the world of molecular dynamics, so we instinctively shy away from reactive chemistry in models, but there’s no reason that a generative model like Sesame need be constrained in this manner - just a question of encoding those stimuli.
@johnhanacek@NTallapragada 2/ One solution is iterative: generate against a frozen pocket, use compchem tools to predict the pocket structure in the bound complex, generate against the new structure, and repeat. No guarantee if or how fast it’ll converge.
@johnhanacek@NTallapragada 1/ There’s two levels at which this applies: induced fit, and disordered proteins. The second is blue sky for us right now - we want to talk to people who do this - but induced fit is very much on our minds!
@RolandDunbrack@NTallapragada Not yet, but very much part of our plan to validate predictions & improve models
That said, Sesame is a source of ideas, whether or not you synthesize its outputs. Medchemists can use these to improve molecules they’re already making (eg with Sesame's scaffold-based generation)
1/ Open, Sesame ⚗️
This week, we released Sesame: the first generative chemistry model that fills any protein pocket with drug-like molecules – de novo or from a scaffold – with no constraints on size or class.