Seminars

In a post-"AlphaFold has solved structure prediction" world, our lab is obsessed with the concept of statistical structural biology. First, we collect large datasets (X-ray fragment screens from 1000s of individual crystals) and use new statistical approaches to identify small molecule binders. This inspires new inhibitors, allosteric modulators, and enzyme design strategies. Second, we examine how experimental information in X-ray crystallography and CryoEM encodes statistical distributions of conformations. This inspires software (e.g. qFit) that reveals hidden conformations and new guidance frameworks for diffusion models. Our work reveals the extent of memorization in current models and suggests experiments to extract even more information for improved training. These two statistical approaches to structural biology are synergistic in examining many aspects of biological mechanism. A current focus is the promiscuity of ligand binding in drug metabolizing proteins, as part of the OpenADMET consortium.