GraphBind

GraphBind predicts nucleic-acid-binding residues on proteins using hierarchical graph neural networks (HGNNs) to characterize protein–nucleic acid interactions.

Key Features:

• Graph Neural Network Architecture: Employs an end-to-end graph neural network framework that models proteins as graphs constructed from the spatial neighborhood and local tertiary structures surrounding target residues.
• Hierarchical Embedding: Utilizes hierarchical graph neural networks (HGNNs) to embed structural and bio-physicochemical characteristics of residues.
• Benchmarking and Evaluation: Evaluated on benchmark datasets from the BioLiP database with training and test sets split by release dates for validation on nucleic acid-binding proteins.
• Comparative Analysis: Compared against state-of-the-art methods for predicting small ligand-binding residues, including Ca2+, Mn2+, Mg2+, ATP, and HEME.
• Generalization Capability: Extends the predictive framework to other ligand-binding residue predictions, demonstrating cross-ligand generalization.

Scientific Applications:

• Protein–nucleic acid interaction mapping: Facilitates identification of nucleic-acid-binding residues to support molecular-level studies of protein–nucleic acid interactions.
• Rational drug design: Supports rational design of therapeutics targeting protein–nucleic acid interfaces by pinpointing residue-level binding sites.
• Ligand-binding prediction: Applicable to predicting residues binding small ligands such as Ca2+, Mn2+, Mg2+, ATP, and HEME.

Methodology:

Implements an end-to-end graph neural network framework that constructs graphs from spatial neighborhoods and local tertiary structures, applies hierarchical GNN embedding of structural and bio-physicochemical features, and is benchmarked using BioLiP datasets split by release dates with comparative evaluation against methods for Ca2+, Mn2+, Mg2+, ATP, and HEME-binding residue prediction.