HotSprint: database of computational hot spots in protein interfaces

Area of Science:

• Biochemistry
• Structural Biology
• Bioinformatics

Background:

• Protein-protein interactions are crucial for biological processes.
• Identifying key residues in protein interfaces is essential for understanding binding mechanisms.
• Hot spots are a small subset of interface residues that contribute disproportionately to binding energy.

Purpose of the Study:

• To develop and present HotSprint, a comprehensive database of computational hot spots in protein interfaces.
• To provide a resource for identifying functionally and structurally important residues in protein-protein interactions.
• To evaluate the accuracy of computational hot spot prediction compared to experimental data and machine learning methods.

Main Methods:

• Extraction of 49,512 multi-chain protein structures from the Protein Data Bank (PDB) as of February 2006.
• Identification of conserved residues within interfaces meeting specific buried accessible solvent area (ASA) and complex ASA thresholds as computational hot spots.
• Application of Support Vector Machines (SVM), Decision Trees, and Decision Lists for comparative hot spot prediction.

Main Results:

• The HotSprint database contains data for 35,776 protein interfaces.
• Computational hot spots predicted by the empirical approach correlated with experimental hot spots with 76% accuracy.
• The empirical approach demonstrated superior performance compared to SVM, Decision Trees, and Decision Lists for hot spot prediction.

Conclusions:

• HotSprint serves as a valuable resource for researchers studying protein interfaces.
• The database provides insights into the evolutionary history and solvent accessibility of interface residues.
• The developed computational method offers an accurate and efficient way to predict functionally important residues in protein interfaces.