AlphaFold2 Model Refinement Using Structure Decoys.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• AlphaFold2 predictions are crucial for interpreting cryo-electron microscopy (cryo-EM) maps in structural biology.
• Accuracy can be limited by prediction quality, experimental data, and model-map alignment.
• Refining AlphaFold2 models with experimental data is essential for reliable structural insights.

Purpose of the Study:

• To develop and evaluate a novel refinement strategy for AlphaFold2 models using cryo-EM data.
• To assess the effectiveness of generating structural decoys to improve model accuracy.
• To compare different decoy generation methods, 3DRobot and Elastic Network Model (ENM)-based ModeHunter.

Main Methods:

• Generated unbiased structural decoys from an AlphaFold2 model using 3DRobot and ENM-based ModeHunter.
• Applied these decoys to refine a challenging AlphaFold2 model against a cryo-EM map of a lipid-preserved respiratory supercomplex.
• Evaluated decoy set performance by selecting the decoy that best fits the experimental cryo-EM density (TM-score).

Main Results:

• 3DRobot produced compact decoys, offering minimal improvement to the initial AlphaFold2 model (best decoy TM-score: 0.53).
• ENM-based ModeHunter generated extended decoys, significantly enhancing the accuracy of the AlphaFold2 model (best decoy TM-score: 0.68).
• ENM demonstrated superior performance in refining the AlphaFold2 model against the cryo-EM map compared to 3DRobot.

Conclusions:

• ENM-based decoy generation is a promising approach for refining AlphaFold2 models using cryo-EM data.
• This method effectively addresses limitations in AlphaFold2 predictions for complex biological structures.
• The strategy of selecting the best-fitting decoy significantly improves structural model accuracy in cryo-EM map interpretation.