Connecting the dots: deep learning-based automated model building methods in cryo-EM.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Single particle cryo-electron microscopy (cryo-EM) has advanced biomolecular complex structure determination.
• Accurate atomic model building from cryo-EM density maps is crucial but challenging at both low and high resolutions.

Purpose of the Study:

• To review deep learning-based methods for automating atomic model building in cryo-EM density maps.
• To assess the impact of these methods on streamlining structure determination.
• To classify deep learning tools based on the hierarchical organization of biomolecular structures they model.

Main Methods:

• Classification of deep learning methods based on modeling primary, secondary, tertiary, and quaternary structures.
• Categorization of tools into 'de novo' (direct prediction) and 'hybrid' (template integration) approaches.
• Discussion of limitations including dataset size, diversity, and conformational heterogeneity.

Main Results:

• Deep learning offers powerful tools to automate and accelerate atomic model building in cryo-EM.
• Methods are categorized by their ability to model different levels of biomolecular structure.
• Both de novo and hybrid deep learning approaches are being developed.

Conclusions:

• Deep learning significantly streamlines structure determination in cryo-EM.
• Challenges remain in dataset curation and handling conformational heterogeneity.
• Future directions promise automated, data-driven model building as a standard in structural biology.