JAMUN: Bridging Smoothed Molecular Dynamics and Score-Based Learning for Conformational Ensemble Generation

Conformational ensembles of protein structures are immensely important both to understanding protein function, and for drug discovery in novel modalities such as cryptic pockets. Current techniques for sampling ensembles such as molecular dynamics are computationally inefficient. On the other hand, many recent machine-learning methods do not generalize well outside their training data. We propose JAMUN, that performs MD in a smoothed, noised space of all-atom 3D conformations of molecules by utilizing the framework of walk-jump sampling. JAMUN enables ensemble generation at orders of magnitude faster rates than traditional molecular dynamics or state-of-the-art ML methods. This physical prior enables JAMUN to transfer to systems outside its training data. JAMUN is even able to generalize across length scales it was not trained on.