DockNet

DockNet employs a Siamese graph-based neural network to predict contact residues between interacting proteins for identification of protein-protein interaction (PPI) sites in applications such as drug discovery.

Key Features:

• Siamese Graph-Based Neural Network Architecture: A paired graph neural network architecture processes two protein graphs to predict inter-protein contact residues without treating proteins as rigid bodies.
• Comprehensive Residue-Level Node Features: Inputs include residue type, surface accessibility, residue depth, secondary structure, pharmacophore properties, and torsional angles at the residue level.
• Modeling of Protein Flexibility: The method incorporates entire protein structures to accommodate protein flexibility during interactions rather than relying on rigid-body assumptions.
• Performance on Benchmark Dataset: Demonstrated performance includes an area under the curve (AUC) of up to 0.84 on the independent test set DB5 and applicability when accurate unbound structures are unavailable.

Scientific Applications:

• Drug discovery and design: Prediction of PPI contact residues to guide identification of interaction interfaces for small-molecule or biologic intervention.
• Target identification and modulation of PPIs: Identification of candidate interface residues for designing modulators that disrupt or stabilize protein-protein interactions.

Methodology:

DockNet uses a Siamese graph-based neural network that analyzes entire protein structures by processing residue-level node features (residue type, surface accessibility, residue depth, secondary structure, pharmacophore properties, torsional angles) to predict inter-protein contact residues.