GRaSP-py

GRaSP-py predicts protein–ligand-binding site residues using a graph-based residue neighborhood strategy and supervised learning to model atomic-level residue environments for functional and interaction inference.

Key Features:

• Graph-based residue neighborhood: Represents residue environments as graphs to capture spatial and relational information between atoms and residues.
• Atomic-level modeling: Models residue environments at the atomic level to incorporate fine-grained structural details relevant to binding.
• Supervised learning: Uses supervised machine learning to learn patterns of binding-site residues from annotated data.
• Residue-centric prediction: Predicts binding-site residues directly rather than focusing on pocket-centric detection.
• Captures complex interactions: Employs graph-theoretical features to capture complex intra-structural interactions within protein structures.
• Benchmark performance: Outperforms six other residue-centric methods including state-of-the-art approaches and surpasses the method evaluated in CAMEO, while ranking second against five leading pocket-centric methods.
• Scalability and efficiency: Predicts binding sites for a protein complex in an average of 10–20 seconds, compared with 2–5 hours reported for a state-of-the-art residue-centric method.

Scientific Applications:

• Protein function annotation: Identifies ligand-binding residues to support functional inference of proteins.
• Discovery of novel functional roles: Detects potential ligand-binding residues to hypothesize new activities or interaction partners.
• Large-scale screening: Enables rapid prediction of binding residues across many protein structures for high-throughput studies.
• Method comparison and evaluation: Serves as a residue-centric benchmark against pocket-centric and other residue-centric prediction methods.

Methodology:

Represents residue neighborhoods as graphs at atomic resolution and applies supervised learning to predict binding-site residues.