RDOCK: refinement of rigid-body protein docking predictions

Area of Science:

• Computational Biology
• Structural Biology
• Biochemistry

Background:

• Protein-protein interactions are crucial for biological processes.
• Accurate prediction of protein complex structures is essential for understanding function.
• Existing rigid-body docking methods like ZDOCK provide initial predictions but require refinement.

Purpose of the Study:

• To present RDOCK, a novel and simple algorithm for refining unbound protein-protein docking predictions.
• To evaluate the effectiveness of RDOCK in improving the accuracy of ZDOCK predictions.
• To assess the performance of RDOCK across a diverse benchmark of protein complexes.

Main Methods:

• RDOCK employs a three-stage energy minimization scheme.
• Electrostatic and desolvation energies are evaluated during refinement.
• Ionic side chains are neutralized in initial minimization stages and re-charged later.

Main Results:

• RDOCK substantially improves ZDOCK predictions across all tested scoring functions and benchmark categories.
• The combination of RDOCK with ZDOCK2.1 (pairwise shape complementarity scoring) achieved top-4 predictions for 49% of cases.
• Binding funnels were observed for 24 cases, representing the first report of such phenomena from global searches in a broad benchmark.

Conclusions:

• RDOCK offers a simple yet effective approach to refine unbound docking predictions.
• The automated RDOCK algorithm significantly enhances the accuracy of protein complex structure prediction.
• RDOCK, particularly when combined with ZDOCK, represents a powerful tool for computational structural biology.