DiffModeler: large macromolecular structure modeling for cryo-EM maps using a diffusion model.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Cryo-electron microscopy (cryo-EM) is crucial for determining the structure of large protein complexes.
• Modeling large complexes (>10 chains) from cryo-EM data is challenging, especially at lower resolutions.

Purpose of the Study:

• To present DiffModeler, a fully automated computational method for modeling large protein complex structures from cryo-EM maps.
• To demonstrate the effectiveness of DiffModeler across various resolutions, including low-resolution data.

Main Methods:

• DiffModeler utilizes a diffusion model for accurate backbone tracing.
• It integrates AlphaFold2-predicted single-chain structures for precise model fitting.
• The method was evaluated on cryo-EM datasets with resolutions ranging from 0-20 Å.

Main Results:

• DiffModeler achieved high template modeling scores: 0.88 and 0.91 for 0-5 Å maps, and 0.92 for 5-10 Å maps.
• The method demonstrated substantially improved performance compared to existing techniques.
• Versatile performance was confirmed at low resolutions (10-20 Å), showing plausible results.

Conclusions:

• DiffModeler provides a robust and automated solution for modeling large protein complexes from cryo-EM data.
• Its ability to perform well across a range of resolutions, including low-resolution maps, enhances its utility in structural biology.
• This method advances the field of structural determination for complex biological assemblies.