An automated decision-tree approach to predicting protein interaction hot spots

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Protein-protein interactions (PPIs) are vital in biological processes.
• Hot spots are key residues driving PPI binding energy.
• Experimental identification of hot spots is resource-intensive.

Purpose of the Study:

• To develop and validate novel computational models for predicting PPI hot spots.
• To assess the predictive accuracy of new models against existing methods.
• To apply the models to biologically relevant protein interfaces.

Main Methods:

• Development of two knowledge-based models: K-FADE (using shape specificity) and K-CON (using biochemical contacts).
• Combination of K-FADE and K-CON into the KFC model.
• Comparison of KFC and combined KFC/Robetta-Ala models against computational alanine scanning.
• Application of KFC to calmodulin (CaM)/smMLCK and BMP-2/BMPR-IA interfaces.

Main Results:

• The KFC model demonstrates improved predictive accuracy over computational alanine scanning (Robetta-Ala).
• Combining KFC with Robetta-Ala significantly enhances overall hot spot prediction accuracy.
• KFC hot spot predictions correlate strongly with mutations affecting binding affinity in tested interfaces.

Conclusions:

• K-FADE and K-CON provide accurate computational predictions of protein-protein interaction hot spots.
• The combined KFC model and its integration with other methods offer a powerful tool for identifying critical residues in PPIs.
• Accurate hot spot prediction can guide experimental efforts and facilitate the design of molecules modulating PPIs.