Improved Denoising of Cryo-Electron Microscopy Micrographs with Simulation-Aware Pretraining.

Area of Science:

• Structural biology
• Biophysics
• Microscopy

Background:

• Cryo-electron microscopy (cryo-EM) is crucial for determining biological macromolecule structures.
• Cryo-EM images suffer from low signal-to-noise ratio (SNR) due to imaging constraints, hindering analysis.
• Existing supervised denoising methods struggle with experimental data due to a lack of ground truth datasets.

Purpose of the Study:

• To develop an efficient denoising algorithm for cryo-EM micrographs.
• To improve the signal-to-noise ratio (SNR) of cryo-EM images for enhanced macromolecular analysis.
• To create a denoising model that generalizes well to experimental cryo-EM data.

Main Methods:

• Introduced a simulation-aware image denoising (SaID) pretrained model.
• Developed a parameter calibration algorithm for generating accurate simulated cryo-EM datasets.
• Trained a deep denoising model using a precisely simulated dataset to ensure generalization.

Main Results:

• The SaID model significantly enhances the SNR of cryo-EM micrographs.
• The denoising performance on experimental cryo-EM data was excellent.
• The method effectively streamlines downstream analyses in cryo-EM studies.

Conclusions:

• The simulation-aware denoising approach provides a robust solution for low-SNR cryo-EM images.
• Accurate data simulation is key to training generalizable denoising models.
• This technique advances the efficiency and efficacy of structural determination using cryo-EM.