Reliable protein-protein docking with AlphaFold, Rosetta, and replica-exchange

Area of Science:

• Computational biology
• Structural biology
• Biophysics

Background:

• Predicting protein complex structures is crucial for understanding biological functions.
• AlphaFold (AF) and AF-multimer (AFm) have advanced protein structure prediction but struggle with conformational changes in complexes.
• Existing methods achieve limited success rates, particularly for dynamic protein interactions.

Purpose of the Study:

• To develop an improved computational pipeline for accurate protein complex structure prediction.
• To address limitations of deep learning methods in modeling conformational flexibility during docking.
• To enhance the prediction accuracy for challenging protein-protein interactions.

Main Methods:

• Combined AlphaFold for structural template generation with a physics-based replica exchange docking algorithm (ReplicaDock 2.0).
• Utilized AlphaFold's predicted Local Distance Difference Test (pLDDT) scores to estimate protein flexibility and docking accuracy.
• Developed AlphaRED (AlphaFold-initiated Replica Exchange Docking) pipeline for enhanced sampling of conformational changes.

Main Results:

• AlphaRED successfully docked 97 previously failed AlphaFold predictions from Docking Benchmark Set 5.5.
• Achieved CAPRI acceptable-quality or better predictions for 63% of benchmark targets.
• Demonstrated a 43% success rate on difficult antigen-antibody targets, significantly outperforming AFm's 20% success rate.

Conclusions:

• Integrating deep learning (AlphaFold) with physics-based methods (ReplicaDock 2.0) provides a robust strategy for protein complex structure prediction.
• AlphaRED effectively models conformational changes, improving accuracy for challenging protein interfaces.
• The AlphaRED pipeline offers a significant advancement in computational structural biology for predicting dynamic protein interactions.