Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust single-particle cryo-EM image denoising and restoration

Jing Zhang, Tengfei Zhao, ShiYu Hu, Xin Zhao

TL;DR

This work tackles the challenge of denoising single-particle cryo-EM images plagued by extremely low SNR and complex structural noise. It introduces a conditional diffusion framework, SR3-based, to estimate $p_ heta(y_0|x)$ with a noise-level conditioning, followed by a lightweight post-processing module to accelerate convergence. The approach demonstrates superior denoising performance on simulated and real datasets and yields higher-fidelity 3D reconstructions, as evidenced by FSC 0.143-based resolutions and improved metrics. The method offers practical impact by enabling clearer particle structures and more accurate density maps, facilitating downstream structural analysis.

Abstract

Cryo-electron microscopy (cryo-EM) has achieved near-atomic level resolution of biomolecules by reconstructing 2D micrographs. However, the resolution and accuracy of the reconstructed particles are significantly reduced due to the extremely low signal-to-noise ratio (SNR) and complex noise structure of cryo-EM images. In this paper, we introduce a diffusion model with post-processing framework to effectively denoise and restore single particle cryo-EM images. Our method outperforms the state-of-the-art (SOTA) denoising methods by effectively removing structural noise that has not been addressed before. Additionally, more accurate and high-resolution three-dimensional reconstruction structures can be obtained from denoised cryo-EM images.

Robust single-particle cryo-EM image denoising and restoration

TL;DR

This work tackles the challenge of denoising single-particle cryo-EM images plagued by extremely low SNR and complex structural noise. It introduces a conditional diffusion framework, SR3-based, to estimate with a noise-level conditioning, followed by a lightweight post-processing module to accelerate convergence. The approach demonstrates superior denoising performance on simulated and real datasets and yields higher-fidelity 3D reconstructions, as evidenced by FSC 0.143-based resolutions and improved metrics. The method offers practical impact by enabling clearer particle structures and more accurate density maps, facilitating downstream structural analysis.

Abstract

Cryo-electron microscopy (cryo-EM) has achieved near-atomic level resolution of biomolecules by reconstructing 2D micrographs. However, the resolution and accuracy of the reconstructed particles are significantly reduced due to the extremely low signal-to-noise ratio (SNR) and complex noise structure of cryo-EM images. In this paper, we introduce a diffusion model with post-processing framework to effectively denoise and restore single particle cryo-EM images. Our method outperforms the state-of-the-art (SOTA) denoising methods by effectively removing structural noise that has not been addressed before. Additionally, more accurate and high-resolution three-dimensional reconstruction structures can be obtained from denoised cryo-EM images.
Paper Structure (13 sections, 3 equations, 5 figures, 1 table)

This paper contains 13 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. METHOD
  3. Architecture Design
  4. Diffusion Model
  5. Post-processing Module
  6. Training
  7. Experiment
  8. Datasets
  9. Experimental Setup
  10. Experimental Scheme
  11. Evaluation Metrics
  12. Experimental results
  13. CONCLUSIONS

Figures (5)

  • Figure 1: Noises cryo-EM images of the HCN1 channellee2017structures. (a) Real cryo-EM image (EMPIAR-10081). (b) 3D cryo-EM density map (EMD-8511, yellow box represents structural noise). (c) Biological structure (PDB: 5u6o).
  • Figure 2: Framework Overview. It includes two parts: diffusion model and post-processing module. The diffusion model restores the particle structure, and the post-processing accelerates the convergence.
  • Figure 3: Comparison of the denoising results from different methods. The first and second rows compare the denoising results of the EMD-8511 and EMD-24928 simulated datasets respectively, and the third row compares the denoising results in the real dataset of the EMPAIR-10081, whose ground truth is the closest projection map. Our method can effectively remove unresolved structural noise on both simulated and real datasets under high noise conditions.
  • Figure 4: The 3D reconstruction results of denoised images. Our results are closest to the ground truth.
  • Figure 5: FSC of the 3D reconstruction results. According to the 0.143 cutoff criterion (Dashed line), as the spatial frequency increases, the resolution of 3D structures gradually increases and the FSC value of highly reliable structures steadily approaches 0. Our results exhibits highly reliability and high-resolution.