Abstract 2100 UNCOVERING THE PROTEIN STRUCTURAL ENSEMBLE BY INTEGRATING MOLECULAR MODELING, ARTIFICIAL INTELLIGENCE, AND CRYO-EM

Cryogenic-sample electron microscopy (cryo-EM) has become a leading technique for determining the structure of biological macromolecules. However, many biological molecules are structurally heterogeneous, occupying a broad range of possible conformations. In principle, cryo-EM gives us the tools to extract this heterogeneity as well: biomolecules are trapped in the vitreous water in conformations close to the ones they adopt in solution. Our aim is to develop new algorithms that leverage our knowledge of protein biophysics to extract this information. To this aim, we present an approach towards analyzing cryo-EM experiments by comparing individual particle images with structures genereted through physical simulation. Initial results suggest that this approach can quantitatively recover the conformational probability distribution of a biomolecule from cryo-EM data. We then build on this prior work by introducing new algorithms that can handle more diverse structural ensembles. First, we present algorithmic improvements that allow us to efficiently determine which image-structure comparisons are most meaningful, considerably improving the speed of our algorithm. Secondly, we introduce a feedback loop where cryo-EM images directly influence the molecular dynamics simulations used. Together, these algorithmic developments have the potential to unlock new experimental approaches for uncovering the conformational distribution of highly flexible biomolecules.