Protein model refinement for cryo-EM maps using AlphaFold2 and the DAQ score.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Cryo-electron microscopy (cryo-EM) is increasingly used to determine protein structures.
• Ensuring the accuracy of these protein models is critical for structural biology research and data deposition.
• Current methods for evaluating and correcting errors in cryo-EM models require improvement.

Purpose of the Study:

• To develop and validate a novel protocol for assessing the quality of protein models derived from cryo-EM data.
• To implement a local structure refinement strategy for correcting errors in these models.
• To compare the efficacy of the new protocol against existing methods.

Main Methods:

• Utilized a deep-learning-based score, DAQ, for model-local map assessment.
• Employed a modified AlphaFold2 procedure for local structure refinement, using trimmed inputs to control refinement regions.
• Conducted benchmark studies to evaluate the protocol's performance on various protein models.

Main Results:

• The DAQ-refine protocol consistently improved low-quality regions in initial protein models.
• The DAQ score demonstrated a strong correlation with actual model quality.
• The protocol successfully identified the most accurate model among refined options in most test cases.
• Improvements from DAQ-refine were, on average, superior to those from other established methods.

Conclusions:

• The DAQ-refine protocol offers a robust approach for evaluating and enhancing protein models generated from cryo-EM density maps.
• This method contributes to improving the reliability of structural models deposited in public databases like the PDB.
• The DAQ score serves as an effective metric for assessing model accuracy and guiding refinement efforts.