Densest subgraph-based methods for protein-protein interaction hot spot prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Network Science

Background:

• Hot spots are crucial for protein binding analysis.
• Residue interaction networks are key for hot spot prediction.
• Existing graph theory methods suffer from low recall.

Purpose of the Study:

• To develop novel graph theory-based methods for predicting protein hot spots.
• To improve the recall and F-score of hot spot prediction using residue interaction networks.
• To identify potential hot spots more effectively from a single network.

Main Methods:

• Developed three graph theory-based methods to predict hot spots.
• Identified important residues by detecting subgraphs with high average degrees (high density).
• Utilized residue interaction networks constructed from spatial atomic coordinate data.

Main Results:

• The developed methods, particularly Min-SDS, significantly outperform existing graph theory methods.
• Min-SDS achieved an average recall of over 0.665 and an f2-score of over 0.364.
• Existing methods had recall and f2-scores below 0.400 and 0.224, respectively.

Conclusions:

• The Min-SDS method demonstrates superior performance in hot spot prediction.
• The proposed methods offer effective approaches for analyzing biological networks.
• Densest subgraph-based methods predict hot spots using single residue interaction networks, reducing reliance on experimental data.