PROCOS: computational analysis of protein-protein complexes

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biochemistry

Background:

• Evaluating the accuracy of protein-protein docking solutions is a significant challenge in structural biology.
• Existing scoring functions often lack a standardized scale, hindering cross-target comparisons.
• There is a need for robust, automated methods to assess the quality of predicted protein complexes.

Purpose of the Study:

• To introduce PROCOS, a novel program for calculating a probability-like measure of native status for protein complexes.
• To provide a scoring system with a fixed range, enabling comparisons across different protein targets.
• To develop a reliable, automated method for evaluating protein-protein complex structures.

Main Methods:

• PROCOS utilizes property distributions derived from a large database of native and non-native protein complexes.
• A support vector machine (SVM) is employed for complex analysis, integrating these property distributions.
• The performance of PROCOS was benchmarked against established scoring schemes like ZRANK and DFIRE.

Main Results:

• PROCOS achieved high probability values (>50%) for 90% of experimentally validated native complexes.
• Testing on decoy sets (Dockground, RosettaDock) and CAPRI targets demonstrated PROCOS's effectiveness.
• The probability-based scoring system successfully identified a significant number of near-native complexes within top rankings.

Conclusions:

• PROCOS presents a novel, fully automated method for the reliable evaluation of protein-protein complexes.
• The probability-based scoring offers advantages for comparing models across different targets and algorithms.
• This approach enhances the assessment of protein-protein docking predictions.