DeepEMhancer: a deep learning solution for cryo-EM volume post-processing

Abstract

Cryo-electron microscopy (cryo-EM) maps are among the most valuable sources of information for protein structure modeling. However, due to the loss of contrast at high frequencies, they generally need to be post-processed before modeling in order to improve their interpretability. To that end, approaches based on B-factor correction are the most popular choices, yet they suffer from some limitations such as the fact that the correction is applied globally, ignoring the presence of heterogeneity in the map local quality that cryo-EM reconstructions tend to exhibit. With the aim of overcoming these limitations, here we present DeepEMhancer, a deep learning approach designed to perform automatic post-processing of cryo-EM maps. Trained on a dataset of pairs of experimental cryo-EM maps and maps sharpened by LocScape using their respective atomic models, DeepEMhancer has automatically learned how to post-process experimental maps performing masking-like and sharpening-like operations in a single step. DeepEMhancer has been evaluated on a testing set of 20 different experimental maps, showing its ability to obtain much cleaner and detailed versions of the experimental maps, thus, improving their interpretability. Additionally, we have illustrated the benefits of DeepEMhancer with a use case in which the structure of the SARS-CoV 2 RNA polymerase is improved.

Continuous flexibility analysis of SARS-CoV-2 Spike prefusion structures

“…The basic elements of the workflow combine quite classic cryo-EM algorithms with recent improvements in particle picking ( Sanchez-Garcia et al, 2020b , 2018 ; Wagner et al, 2019 ) and key ideas of meta classifiers, which integrate multiple classifiers by a “consensus” approach ( Sorzano et al, 2000 ), finalizing with a totally new approach to map post-processing based on deep learning that we term “Deep cryo EM Map Enhancer” ( Sanchez-Garcia et al, 2020a ), that complements our previous proposal on local deblurring ( Ramírez-Aportela et al, 2020b ). Naturally, map and map-model quality analysis are performed with a variety of tools ( Pintilie et al, 2020 ; Ramírez-Aportela et al, 2020a ; Vilas et al, 2020 ).…”

“…One of the approaches is our already introduced LocalDeblur sharpening method (Ramírez-Aportela et al, 2020b). The second approach is a totally new method based on deep learning (Sanchez-Garcia et al, 2020a). Concentrating on the latter method, DeepEMhancer, it relies on a common approach in modern pattern recognition, where a Convolutional Neural Network (CNN) is trained on a known data set, comprised of pairs of data points and targets, with the aim of predicting the targets for new unseen data points.…”

Continuous flexibility analysis of SARS-CoV-2 spike prefusion structures

“…The basic elements of the workflow combine classic cryo-EM algorithms with recent improvements in particle picking (Sanchez-Garcia et al, 2018;Sanchez-Garcia, Segura et al, 2020;Wagner et al, 2019) and the key ideas of meta classifiers, which integrate multiple classifiers by a 'consensus' approach (Sorzano et al, 2020), and finish with a totally new approach to map post-processing based on deep learning that we term Deep cryo-EM Map Enhancer (DeepEMhancer; Sanchez-Garcia, Gomez-Blanco et al, 2020), which complements our previous proposal on local deblurring (Ramírez-Aportela, Vilas et al, 2020). Naturally, map and map-model quality analyses are performed using a variety of tools (Pintilie et al, 2020;Vilas et al, 2020).…”

Mechanisms of antibody binding revealed by asymmetric Fab-virus complexes

“…This matching approach provided equal data sets for a total of 1,657,372 particleorientations per reconstruction. DeepEMhancer was used to improve local sharpening of maps during post-processing (33). Difference maps were calculated by subtracting the Fab-occupied and unoccupied maps in EMAN (34).…”