Combining interface core and whole interface descriptors in postscan processing of protein-protein docking models

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• Protein-protein docking simulations often generate numerous models, with correct interactions sometimes poorly ranked.
• Distinguishing accurate models from false positives remains a challenge in computational structural biology.

Purpose of the Study:

• To develop and validate postscan processing procedures to enhance the accuracy of protein-protein docking predictions.
• To improve the ranking and selection of nearly correct models (NCMs) from docking scans.

Main Methods:

• Utilized propensity descriptors, interface-core cluster counts, solvation energy, and complementarity scores for postscan analysis.
• Implemented soft intersection filters using Boolean classifiers to eliminate false models.
• Developed new scoring functions incorporating standardized descriptors to highlight NCMs.
• Tested procedures on unbound docking models generated by MolFit without external data.

Main Results:

• Postscan filters significantly reduced the number of putative models across different system classes (e.g., enzyme-inhibitor, antibody-antigen).
• Combined filters and scoring functions improved the average rank of top-ranked NCMs from 673 to 122.
• Effectiveness demonstrated on CAPRI targets, including those using external information.
• Class-specific discrimination between correct and false models was observed.

Conclusions:

• The developed postscan procedures effectively enhance the distinction between nearly correct and false protein-protein interaction models.
• Class-specific approaches are crucial for optimizing the performance of these postscan methods.
• The findings offer a significant advancement in refining computational docking predictions for structural biology research.