Edge-aware GAT-based protein binding sites prediction

Area of Science:

• Computational biology
• Structural bioinformatics
• Drug discovery

Background:

• Accurate identification of protein binding sites is essential for understanding biomolecular interactions and designing targeted drugs.
• Existing predictive methods face challenges in balancing accuracy and computational efficiency for complex spatial structures.

Purpose of the Study:

• To introduce an Edge-aware Graph Attention Network (Edge-aware GAT) for fine-grained binding site prediction across proteins, DNA/RNA, ions, ligands, and lipids.
• To improve the accuracy and efficiency of binding site identification using a structure-based approach.

Main Methods:

• Constructing atom-level graphs integrating geometric descriptors, DSSP secondary structure, and relative solvent accessibility (RSA).
• Utilizing interatomic distances and unit direction vectors as edge features within an attention mechanism for enhanced local structural representation.
• Employing geometry-aware edge attention and residue-level attention pooling.

Main Results:

• Achieved a ROC-AUC of 0.93 for protein-protein binding site prediction on benchmark datasets.
• Demonstrated competitive performance compared to other structure-based atomic-level models.
• Confirmed practical utility and interpretability through PyMOL visualizations.

Conclusions:

• The Edge-aware GAT provides an efficient, structure-based framework for identifying functional sites in proteins.
• The model balances prediction accuracy, generalization, and interpretability, facilitating drug target design.
• A publicly accessible web server is available for community use.