Restraint Quality, Not Quantity, Predicts Peptide-Protein Docking Outcomes.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Protein-peptide interactions are vital for cellular signaling and drug discovery.
• Traditional structure determination methods (NMR, X-ray crystallography) are time-consuming.
• Current computational methods (docking, deep learning) face challenges with flexible peptides and low sequence identity.

Purpose of the Study:

• To develop a novel restraint scoring function for evaluating the informativeness of distance restraints in protein-peptide docking.
• To introduce a minimal-restraint docking strategy for optimizing restraint subsets and improving structural model quality.
• To provide a scalable and efficient approach for structure prediction in data-limited contexts.

Main Methods:

• Developed a restraint scoring function integrating evolutionary conservation, spatial proximity, and geometric distribution.
• Implemented a minimal-restraint docking strategy to identify optimal restraint subsets.
• Evaluated the approach on diverse protein-peptide systems, including SH3 and WW domain complexes and PepPCBench cases.

Main Results:

• Model quality consistently improved with increasing restraint score.
• Established domain-specific restraint-score thresholds for accurate model selection in SH3 and WW systems.
• Demonstrated the effectiveness of the minimal-restraint strategy in improving structural model accuracy.

Conclusions:

• The restraint scoring function and minimal-restraint strategy offer a scalable and efficient method for protein-peptide structure prediction.
• This approach provides quantifiable confidence in restraint-informed modeling, particularly in data-limited situations.
• Lays the foundation for data-efficient machine learning-based peptide-protein docking.