Cryo-electron microscopy (cryo-EM) has made it possible to reconstruct molecular structures and movies with unprecedented detail. As machine learning continues to transform structural biology, what are the next frontiers for cryo-EM structure determination? In this talk, I will present our lab’s work on AI-driven methods for analyzing structural heterogeneity in cryo-EM, centered on cryoDRGN, a deep learning framework for heterogeneous reconstruction. A central objective is enabling high-throughput structure determination of challenging systems. I will highlight recent extensions, including cryoDRGN-ET for in situ cryo-electron tomography data, and cryoDRGN-AI for ab initio reconstruction of complex molecular mixtures. I will also introduce CryoBench, a community benchmarking effort and data set suite designed to advance heterogeneous reconstruction methods. Finally, I will discuss integrating large-scale generative models of protein structure as sequence-conditioned priors to enable new capabilities in high-resolution reconstruction of dynamic protein complexes.
Ellen Zhong (Sun,) studied this question.