Refinement of Cryo-EM 3D Maps with Self-Supervised Denoising Model: crefDenoiser

Abstract

AbstractCryogenic electron microscopy (cryo-EM) is a pivotal technique for imaging macromolecular structures. Despite extensive processing of large image sets collected in a cryo-EM experiment to amplify the signal-to-noise ratio, the reconstructed 3D protein density maps are often limited in quality due to residual noise, which in turn affects the accuracy of the macromolecular representation. In this paper, we introduce crefDenoiser, a denoising neural network model designed to enhance the signal in 3D cryo-EM maps produced with standard processing pipelines, beyond the current state of the art. crefDenoiser is trained without the need for ‘clean’, ground-truth target maps. Instead, we employ a custom dataset composed of real noisy protein half-maps sourced from the Electron Microscopy Data Bank repository. Strong model performance is achieved by optimizing for the theoretical noise-free map during self-supervised training. We demonstrate that our model successfully amplifies the signal across a wide variety of protein maps, outperforming a classical map denoiser and a network-based sharpening model. Without biasing the map, the proposed denoising method often leads to improved visibility of protein structural features, including protein domains, secondary structure elements, and amino-acid side chains.