Residue-level error detection in cryoelectron microscopy models.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Building accurate protein models from moderate-resolution (3-5 Å) cryo-electron microscopy (cryo-EM) maps presents significant challenges and is prone to errors.
• Existing methods may struggle to identify localized inaccuracies within these complex structural models.

Purpose of the Study:

• To develop and validate a robust computational tool for identifying and correcting local backbone errors in protein structures modeled into cryo-EM maps.
• To enhance the reliability and accuracy of protein structure determination using cryo-EM data.

Main Methods:

• Developed MEDIC (Model Error Detection in Cryo-EM), a statistical model integrating local fit-to-density metrics with deep learning-derived structural information.
• Validated MEDIC on 28 protein structures subsequently solved at higher resolutions.
• Applied MEDIC to correct errors in deposited cryo-EM structures and identify errors in AlphaFold predictions.

Main Results:

• MEDIC achieved 68% precision and 60% recall in identifying differences between low- and high-resolution structures.
• Successfully corrected over 100 errors in 12 deposited cryo-EM structures.
• Identified errors in 4 refined AlphaFold predictions with 80% precision and 60% recall.

Conclusions:

• MEDIC is a powerful and versatile tool for structural biologists, capable of detecting errors in protein models generated through both manual building and deep learning approaches.
• The model enhances the quality control process for cryo-EM-derived protein structures.
• Facilitates more accurate protein structure refinement and rebuilding, particularly as deep learning predictions become more integrated into workflows.