Abstact

Single-particle analysis by Cryo-electron microscopy (CryoEM) provides direct access to the conformations of macromolecules. Traditional methods assume discrete conformations, while newer algorithms estimate conformational landscapes representing the different structural states a biomolecule explores. This work presents HetSIREN, a deep learning-based method that can fully reconstruct or refine a CryoEM volume in real space based on the structural information summarized in a conformational latent space. HetSIREN is defined as an accurate space-based method that allows spatially focused analysis and the introduction of sinusoidal hypernetworks with proven high analytics capacities. Continuing with innovations, HetSIREN can also refine the images' pose while conditioning the network with additional constraints to yield cleaner high-quality volumes, as well as addressing one of the most confusing issues in heterogeneity analysis, as it is the fact that structural heterogeneity estimations are entangled with pose estimation (and to a lesser extent with CTF estimation) thanks to its decoupling architecture.