PPISHES-an enhanced physicochemical approach for predicting protein interaction sites using graph neural networks

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Accurate prediction of protein interaction sites is vital for understanding biological mechanisms.
• Existing methods often neglect crucial physicochemical properties, hindering prediction performance.

Purpose of the Study:

• To develop an enhanced model for predicting protein interaction sites.
• To improve prediction accuracy by integrating solvent accessible surface area, hydrogen-bonding propensity, and electrostatic potential.

Main Methods:

• Developed a novel prediction model incorporating three key physicochemical features.
• Conducted feature ablation studies to identify critical components.
• Evaluated model performance on obligate and non-obligate protein complexes.

Main Results:

• Achieved significant improvements in Area Under the Precision-Recall Curve (up to 42.8% and 29.3%).
• Outperformed state-of-the-art methods in Recall, Area Under the Curve, and Matthews Correlation Coefficient.
• Demonstrated the importance of physicochemical features for enhanced prediction.

Conclusions:

• The enhanced model provides superior prediction of protein interaction sites.
• Physicochemical properties are critical for improving computational models in this field.
• The model offers a valuable tool for biological research and drug discovery.