CryoDDM: CryoEM denoising diffusion model for heterogeneous conformational reconstruction

Abstract

Heterogeneous protein reconstruction can reveal the relationship between protein dynamics and function, which represents one of the foremost challenges in cryogenic electron microscopy (cryo-EM) single-particle analysis (SPA). However, high-intensity noise results in inaccurate parameter estimation during classification and reconstruction, making it difficult to capture the subtle motions of proteins. Here, we present the Cryo-EM denoising diffusion model (CryoDDM), which denoises images while preserving high-frequency structural information, benefiting conformational heterogeneity classification and reconstruction. The image denoised by CryoDDM can be used to improve downstream analysis, enhance reconstruction resolution, discover new protein conformations, and map out conformational variations. Validated against tree diverse experimental datasets, spanning a proteasome, a membrane protein and a spike protein, our results consistently demonstrate the accuracy and superiority of CryoDDM. By guiding reconstruction with denoised images, CryoDDM enables high-resolution reconstruction of heterogeneous conformations, paving the way to illuminate fundamental questions in structural biology. The CryoDDM project are published at https://github.com/BIT-FuweiLi/CryoDDM.