Enhancing cryo-EM structure prediction with DeepTracer and AlphaFold2 integration

Area of Science:

• Structural biology
• Computational biology
• Biomedical applications

Background:

• Protein structure determination is crucial for biomedical applications like vaccine development.
• Automating protein structure model building from electron density maps is challenging due to limited atomic resolution in most experimental maps.
• Current protein structure prediction tools like AlphaFold2 achieve high accuracy but often require time-consuming manual refinement.

Purpose of the Study:

• To develop an automated method for refining protein structures predicted by AlphaFold2.
• To improve the accuracy and efficiency of protein structure modeling.

Main Methods:

• Developed DeepTracer-Refine, an automated method that refines AlphaFold predicted structures.
• Aligned AlphaFold predicted structures with DeepTracer's modeled structures for refinement.
• Evaluated the method on 39 multi-domain proteins.

Main Results:

• Improved average residue coverage from 78.2% to 90.0%.
• Enhanced average local Distance Difference Test (lDDT) score from 0.67 to 0.71.
• Demonstrated superior performance and faster run-time compared to Phenix's AlphaFold refinement, especially for less precise initial AlphaFold models.

Conclusions:

• DeepTracer-Refine offers an effective automated solution for refining AlphaFold protein structures.
• The method significantly improves model accuracy and coverage, addressing limitations of current prediction and refinement techniques.
• DeepTracer-Refine presents a faster and more accurate alternative for protein structure refinement in computational biology.