GRaSP-web

GRaSP-web predicts protein ligand-binding site residues using GRaSP (Graph-based Residue neighborhood Strategy to Predict binding sites), a graph-based residue neighborhood strategy combined with machine learning to characterize binding regions for structural biology and drug discovery.

Key Features:

• Residue-Centric Graph-Based Methodology: Constructs graphs from neighborhood relationships between amino acid residues to capture spatial and chemical interactions within protein structures.
• Machine Learning Integration: Applies machine learning models trained to predict ligand-binding site residues from residue neighborhood graph representations.
• Performance and Scalability: Outperforms six state-of-the-art residue-centric methods with a Matthews Correlation Coefficient (MCC) of 0.61 and predicts binding sites for protein complexes in 10–20 seconds compared with typical 2–5 hour runtimes of other methods.
• Consistency Across Datasets: Maintains an MCC of 0.61 across bound/unbound structures and a large multi-chain protein dataset (4500 entries), indicating robust performance across datasets.

Scientific Applications:

• Structural biology: Identification of ligand-binding regions to support interpretation of protein structure-function relationships.
• Computational biology: Residue-level binding-site predictions to inform modeling and simulation studies.
• Drug discovery: Prioritization of binding-site residues for ligand design and target characterization.
• High-throughput and large-scale analyses: Scalable prediction of binding sites across large protein datasets and multi-chain complexes.

Methodology:

Constructs residue neighborhood graphs that represent spatial and chemical relationships between amino acids and analyzes those graphs using machine learning models trained on known binding-site data.