CryoARC: Atomic-resolution conformational landscapes of protein assemblies from cryo-EM single particles with evolutionary priors

Abstract

Single-particle cryo-electron microscopy (cryo-EM) reveals structural heterogeneity in macromolecular complexes, but recovering continuous conformational landscapes at high resolution remains challenging. Here, we introduce CryoARC, a deep learning framework that integrates evolutionary sequence representations with cryo-EM particle images to reconstruct continuous conformational ensembles at atomic resolution. CryoARC combines a latent representation of particle heterogeneity with a sequence-conditioned structure decoder inspired by protein structure prediction architectures, enabling direct prediction of particle-specific atomic structures. We further introduce a heterogeneous refinement strategy that aggregates per-particle predictions into a canonical density map, improving reconstruction quality and resolution. We evaluate CryoARC on both synthetic and experimental datasets and show that it recovers continuous conformational landscapes together with coherent atomic models. CryoARC demonstrates how sequence-derived structural priors can be combined with cryo-EM particle images for ensemble-based atomic reconstruction of heterogeneous macromolecular systems. CryoARC is fully open source and available at https://gricad-gitlab.univ-grenoble-alpes.fr/GruLab/CryoARC.