LABind: identifying protein binding ligand-aware sites via learning interactions between ligand and protein

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Drug Discovery

Background:

• Protein-ligand interactions are vital for biological processes and drug development.
• Existing experimental and computational methods face limitations in cost, specificity, and ligand encoding.

Purpose of the Study:

• To develop a structure-based computational method, LABind, for predicting protein binding sites in a ligand-aware manner.
• To address the limitations of current methods in handling diverse ligands and predicting binding interactions.

Main Methods:

• LABind employs a graph transformer architecture to analyze local protein structures and binding patterns.
• A cross-attention mechanism is integrated to learn specific protein-ligand binding characteristics.
• The method was evaluated on three benchmark datasets for binding site prediction.

Main Results:

• LABind demonstrated high effectiveness in predicting binding sites for small molecules and ions.
• The method showed strong generalization capabilities on unseen ligands.
• Further analysis confirmed LABind's ability to integrate ligand information for accurate predictions.

Conclusions:

• LABind offers an effective and ligand-aware structure-based approach for binding site prediction.
• The method shows promise for advancing drug discovery and understanding protein-ligand interactions.
• LABind's applications extend to binding site localization, sequence-based predictions, and molecular docking.