Measurement of atomic scattering factors by cryo-electron microscopy

Determination of specimen structure from cryo-electron microscopy (cryo-EM) experiments relies on an accurate model of the electrostatic potential of the specimen. For biological macromolecules, the potential is strongly influenced by the presence of chemical bonds between atoms, a fact unaccounted for by models of electron scattering that are currently standard in the field. We propose a Bayesian approach to the estimation of atomic scattering factors which incorporates the effect of the molecular environment while remaining fast, interpretable and transferable between molecules. Our algorithm infers atomic scattering factors directly from maps of the electrostatic potential determined by cryo-EM single particle analysis, bypassing the need for computationally-intensive theoretical calculations. The algorithm is used to infer empirical scattering factors from high-resolution reconstructions of catalase enzymes. To illustrate its broad applicability, the algorithm is also applied to a training set of publicly-available cryo-EM data. The empirical scattering factors show improved agreement with a test set of cryo-EM reconstructions. The predictions are further validated by comparison with magnetic susceptibility values of organic compounds, as well as by application to the refinement of atomic models. Significance Statement Understanding the structure of biomolecules is key to explaining their function. Cryo-electron microscopy is a method for reconstructing the electrostatic potential distribution of a biological macro-molecule, a quantity which contains information about atomic positions and the redistribution of charge due to chemical bonding. These factors should be modelled when inferring the structure of the molecule from its electrostatic potential. We develop an improved model of the potential that takes into account chemical bonding while remaining computationally tractable. The parameters of the model are inferred from a selection of cryo-electron microscopy datasets using Bayesian methods.