Title - Mixed dimensional Gaussian Mixture Models to analyze heterogeneity in cryo-EM datasets
Abstract - In the past few years, cryo-EM single-particle imaging has been quite helpful for the reconstruction of 3D protein volumes from 2D cryo-EM images, but these 2D images contain particle snapshots belonging to different conformations and not just one protein structure. Recent developments have given rise to various deep learning models which can be used to explore heterogeneity in the cryo-EM data. One of the recent methods, e2gmm (Chen et al.), uses this powerful scheme for the 3D reconstruction of protein structures and their conformations along with the generation of a small latent space describing the conformational and compositional changes. During the presentation, I will talk about the e2gmm model and its results on different datasets.