Are scoring functions in protein-protein docking ready to predict interactomes? Clues from a novel binding affinity...

Area of Science:

• Structural proteomics
• Computational biology
• Biochemistry

Background:

• Accurate prediction of protein-protein binding affinity is crucial for understanding cellular functions and designing therapeutics.
• Current scoring functions used in protein-protein docking often fail to reliably predict binding affinity.
• Developing a comprehensive benchmark is essential for evaluating and improving these predictive models.

Purpose of the Study:

• To create and validate a protein-protein binding affinity benchmark dataset.
• To assess the performance of commonly used scoring algorithms in predicting binding affinities.
• To identify limitations and guide future improvements in computational prediction of protein-protein interactions.

Main Methods:

• Compiled a benchmark dataset of binding constants (K(d)'s) for 81 protein-protein complexes.
• Evaluated nine standard scoring algorithms and one free-energy prediction algorithm.
• Analyzed correlations between predicted scores and experimentally determined binding affinities, categorizing data by methodology and affinity levels.

Main Results:

• All tested algorithms showed poor correlation between predicted scores and actual binding affinity.
• Categorizing data by experimental methodology and affinity (low, medium, high) revealed emerging correlations.
• Significant correlations were observed within specific data subsets, demonstrating benchmark robustness but also large standard deviations.

Conclusions:

• Existing scoring functions are insufficient for accurate protein-protein binding affinity prediction.
• The developed benchmark is a valuable resource for assessing and advancing computational methods.
• Future efforts must focus on improving scoring functions or developing consensus approaches for reliable affinity prediction.