Machine Learning for Determining Protein Structure and Dynamics from Cryo-EM Images

Explore machine learning techniques for determining protein structure and dynamics from cryo-electron microscopy (cryo-EM) images in this 49-minute conference talk by Ellen Zhong of Princeton University. Delve into the challenges of 3D reconstruction from 2D cryo-EM images due to structural heterogeneity and discover how cryoDRGN, an innovative algorithm, leverages deep neural networks to reconstruct continuous distributions of 3D density maps. Learn about the deep generative model underpinning cryoDRGN, its neural representation of 3D volumes, and the learning algorithm used to optimize this representation from unlabeled 2D cryo-EM images. Gain insights into the practical applications of cryoDRGN in discovering new protein structures and visualizing continuous trajectories of protein motion. Examine various extensions of the method for analyzing dynamic protein complexes and interpreting the learned generative model. Access the open-source cryoDRGN software and explore its potential for advancing cryo-EM research in structural biology.