Improvement Of Cryo-em Maps By Simultaneous Local And Non-local Deep Learning

This article presents a three-dimensional deep learning framework, EMReady, to improve the interpretability of cryo-EM maps. EMReady adopts the three-dimensional Swin-Conv-UNet-based network architecture (SCUNet) which combines the advantages of conventional residual convolution for local modeling, swin (shifted window) transformer for non-local modeling, and multiscale UNet for further enhancement of local and non-local modeling. The network is trained by simultaneously minimizing the local smooth L1 distance and maximizing the non-local structural similarity (SSIM) between processed experimental and simulated target maps. EMReady is extensively evaluated on diverse test sets of primary EM maps and half-maps, and is shown to robustly enhance the quality of cryo-EM maps in terms of various map quality metrics.