xBind: an integrated webserver for large language model-enabled cross-molecular protein binding site prediction

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biology

Background:

• Accurate prediction of protein binding sites is crucial for understanding molecular interactions.
• Existing methods often lack the ability to integrate diverse data types or handle various interaction partners.
• Large language models (LLMs) show promise in capturing complex biological patterns.

Purpose of the Study:

• To develop and present xBind, an interactive webserver for cross-molecular binding-site prediction.
• To leverage LLM embeddings and structural features for enhanced prediction accuracy.
• To provide a user-friendly platform for predicting protein-protein, protein-DNA, and protein-RNA binding sites.

Main Methods:

• Utilized LLM embeddings from the ESM-2 model.
• Integrated sequence- and structure-derived features.
• Employed symmetry-aware deep graph neural networks for prediction.
• Developed a web server with interactive visualizations and customizable options.

Main Results:

• xBind successfully predicts residue-level binding sites for various interaction types.
• The webserver integrates on-the-fly protein structure prediction using AlphaFold.
• Interactive results allow for user-adjustable threshold calibration and visualization.

Conclusions:

• xBind offers a versatile and accessible tool for cross-molecular binding-site prediction.
• The integration of LLMs and graph neural networks advances prediction capabilities.
• The webserver facilitates biological research by providing interpretable and customizable binding site predictions.