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

Abstract

Cryo-EM maps are 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 to improve their interpretability. Most popular approaches, based on global B-factor correction, suffer from limitations. For instance, they ignore the heterogeneity in the map local quality that reconstructions tend to exhibit. Aiming to overcome these problems, 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 maps and maps sharpened using their respective atomic models, DeepEMhancer has learned how to post-process experimental maps performing masking-like and sharpening-like operations in a single step. DeepEMhancer was evaluated on a testing set of 20 different experimental maps, showing its ability to reduce noise levels and obtain more detailed versions of the experimental maps. Additionally, we illustrated the benefits of DeepEMhancer on the structure of the SARS-CoV-2 RNA polymerase.

CUL5-ARIH2 E3-E3 ubiquitin ligase structure reveals cullin-specific NEDD8 activation

“…Focused refinement 40 yielded a 3.4 Å resolution map allowing generation of atomic coordinates showing interactions between neddylated CUL5-RBX2 and ARIH2* (Fig. 2a and Extended Data Figs.…”

“…The experiment was performed twice. b, Interactions between ARIH2* (violet) and neddylated CUL5-RBX2 seen in map from ARIH2*-neddylated CRL5 Vif-CBFβ -A3C complex generated using DeepEMhancer 40 . Connections to the ARIH2* Rcat domain, not visible in the map, are indicated as dotted lines.…”

“…3b,c). Final maps were sharpened using RELION 57 postprocessing or DeepEMhancer 40 . To facilitate model building by improved map quality by local sharpening and noise reduction, two half maps from the final refinement were provided without a mask as input to DeepEMhancer 40 .…”

“…The coordinates for a neddylated CUL5-RBX2-ARIH2* subcomplex of the A3C-bound neddylated CRL5 Vif-CBFβ -ARIH2* E3-E3 were subjected to rebuilding, guided by the map processed with DeepEMhancer, and refined using the 3.4-Å resolution postprocessed map arising from focused refinement (A3C E3-E3 catalytic focused in Extended Data Fig. 3a, EMD-12995) 40 . Initial regions of crystal structures were docked in the focus refined map using Chimera v.1.14 (ref.…”

Constitutive signal bias mediated by the human GHRHR splice variant 1

“…The final density maps of GHRH–SV1–G s and apo SV1–G s were automatically postprocessed using DeepEMhancer ( 70 ) to improve the electron microscopy (EM) map quality before model building. For both structures, the initial GHRH, SV1, and G s heterotrimer models were taken from the GHRH–GHRHR–G s –Nb35 complex ( 18 ) (Protein Data Bank [PDB] number: 7CZ5) and mini-G s heterotrimer was taken from the GPR52–mini-G s complex ( 71 ) (PDB number: 6LI3).…”