GRaSP-web: a machine learning strategy to predict binding sites based on residue neighborhood graphs

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Proteins are vital macromolecules performing functions through molecular interactions at binding sites.
• Accurate identification of protein binding sites is crucial for understanding protein function and advancing drug discovery.
• Experimental methods for binding site identification are often costly and time-consuming.

Purpose of the Study:

• To introduce GRaSP-web, a novel web server for predicting ligand binding sites on proteins.
• To provide a fast, accurate, and scalable computational tool for identifying protein binding regions.

Main Methods:

• Development of GRaSP-web, a web server implementing the Graph-based Residue neighborhood Strategy to Predict binding sites (GRaSP).
• GRaSP utilizes a residue-centric, graph-based machine learning approach to predict potential ligand binding residues.
• The method was evaluated against six state-of-the-art residue-centric methods.

Main Results:

• GRaSP-web demonstrated superior performance with a Matthews Correlation Coefficient (MCC) of 0.61, outperforming six existing methods.
• The server exhibits remarkable scalability, predicting binding sites for protein complexes in 10-20 seconds, a significant improvement over methods taking hours.
• Consistent accuracy was observed across bound/unbound protein structures and a large dataset of multi-chain proteins (4500 entries), maintaining an MCC of 0.61.

Conclusions:

• GRaSP-web offers a highly efficient and accurate solution for predicting protein ligand binding sites.
• The web server's speed and consistency make it a valuable tool for computational drug design and protein function research.
• GRaSP-web is freely accessible at https://grasp.ufv.br, promoting wider research applications.