Higher resolution in cryo-EM by the combination of macromolecular prior knowledge and image-processing tools

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• Single-particle cryo-electron microscopy (cryo-EM) is a key technique for determining 3D structures of biological molecules.
• Recent hardware and software advancements have led to exponential growth in cryo-EM structure determination at medium-high resolutions.
• Existing knowledge from previous cryo-EM studies has not been effectively integrated into new structure reconstruction processes.

Purpose of the Study:

• To explore the utility of the deep learning approach, deepEMhancer, as a regularizer within the RELION refinement process for cryo-EM data.
• To assess deepEMhancer's ability to incorporate prior structural information, reduce noise, enhance signal, and improve isotropy in cryo-EM reconstructions.

Main Methods:

• Integration of deepEMhancer as a regularizer into the RELION refinement workflow.
• Application of the combined approach to the 3D structure determination of biological macromolecules, with a focus on membrane proteins.

Main Results:

• DeepEMhancer effectively acts as a regularizer, introducing prior information to improve cryo-EM reconstructions.
• The method demonstrated significant noise reduction and signal enhancement, leading to improved isotropy.
• These enhancements positively impacted image alignment and reduced overfitting during iterative refinement, particularly for flexible membrane proteins.

Conclusions:

• The combination of deepEMhancer and RELION offers a powerful strategy for enhancing cryo-EM structure determination.
• This deep learning-based regularization is especially beneficial for challenging samples like membrane proteins, which often exhibit high disorder and flexibility.
• The approach facilitates more accurate and reliable 3D structure determination of biological molecules using cryo-EM.