Protein docking refinement by convex underestimation in the low-dimensional subspace of encounter complexes

Area of Science:

• Computational Biology
• Biophysics
• Structural Bioinformatics

Background:

• Protein docking is crucial for understanding molecular interactions and drug discovery.
• Current methods face challenges in accurately refining predicted protein complex conformations.
• Identifying near-native models from diverse conformational ensembles remains a significant hurdle.

Purpose of the Study:

• To develop a novel stochastic global optimization algorithm for refining protein docking predictions.
• To improve the quality of conformational ensembles generated during protein docking.
• To enhance the selection of near-native models from refined ensembles.

Main Methods:

• Utilizing density-based clustering to group protein conformations into distinct binding energy funnels.
• Identifying smooth, low-energy "permissive" subspaces within each cluster.
• Employing convex polynomial underestimation of binding energy functions within these subspaces to bias sampling.
• Iteratively refining sampling and underestimation to generate a high-quality ensemble.

Main Results:

• The proposed algorithm significantly improves the quality of protein complex conformations compared to initial sets.
• Computational results on a benchmark of 224 protein complexes demonstrate the algorithm's effectiveness.
• A novel method was developed to further enhance the ensemble for near-native model selection.

Conclusions:

• The novel stochastic global optimization algorithm effectively refines protein docking predictions.
• The approach enhances the ability to identify biologically relevant protein-protein interactions.
• This work offers a promising advancement for computational drug design and molecular modeling.