Cryo-electron microscopy reconstruction is limited to samples with low conformational heterogeneity, where it is possible to acquire, for each conformation, corresponding particles with enough viewing angles. Moreover, extracting information about the weight (or probability) of each conformation is challenging. To overcome these limitations, we develop an ensemble refinement framework using individual cryo-EM particles. We introduce a prior set of conformations (a representative structural ensemble) that can be obtained for example from molecular simulations, AlphaFold or Rosetta. A Bayesian formalism is then proposed to extract the optimal weight of each conformation given the cryo-EM particle set. We validate the method using synthetic images from molecular simulations of a benchmark biomolecule with several metastable states. The method is able to accurately recover the probability of each state and their associated errors, even for an unfolded state that has high conformational variability.
Copyright. All Rights Reserved.