Topaz-Denoise: general deep denoising models for cryoEM and cryoET.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Cryo-electron microscopy (cryoEM) is crucial for protein structure determination.
• Low signal-to-noise ratio (SNR) in cryoEM images hinders data processing and structure resolution.
• Current methods face challenges with particle orientation and data collection efficiency.

Purpose of the Study:

• To develop and present Topaz-Denoise, a deep learning method for enhancing SNR in cryoEM images and cryoET tomograms.
• To demonstrate the utility of Topaz-Denoise in improving downstream analysis and accelerating data collection.
• To provide a general, pre-trained denoising model applicable to new datasets without retraining.

Main Methods:

• Developed Topaz-Denoise, a deep learning algorithm trained on diverse cryoEM image datasets.
• Applied Topaz-Denoise to denoise cryoEM images and cryoET tomograms.
• Utilized denoised data for 3D single particle structure determination and evaluated low-dose data collection.

Main Results:

• Topaz-Denoise significantly increases SNR in cryoEM images and cryoET tomograms.
• Denoising improved micrograph interpretability, enabling the 3D structure solution of clustered protocadherin.
• Low-dose data collection, facilitated by Topaz-Denoise, enhanced downstream analysis and reduced collection time.

Conclusions:

• Topaz-Denoise is an effective deep learning tool for improving cryoEM and cryoET data quality.
• The method accelerates structure determination and enables lower electron dose data collection.
• Topaz-Denoise is expected to be widely adopted by the cryoEM community for enhanced interpretability and analysis speed.