G-RANK

G-RANK scores protein–protein docking models using an equivariant graph neural network to prioritize near-native protein complex structures for downstream biological engineering applications.

Key Features:

• Equivariant Graph Neural Network (GVP-GNN): Implements a geometric vector perceptron-graph neural network (GVP-GNN) that preserves spatial orientation and symmetry for 3D molecular structure modeling.
• Competitive performance on benchmarks: Evaluated on two distinct test datasets and shown to perform comparable to or better than existing state-of-the-art scoring functions in identifying near-native models.

Scientific Applications:

• Molecular interaction analysis: Prioritizes near-native docking models to improve understanding of protein–protein interfaces.
• Drug discovery: Supports identification of candidate protein–protein interactions relevant to therapeutic targeting.
• Protein engineering and biomaterials: Aids enzyme design and the development of novel biomaterials by improving scoring of docking models.

Methodology:

G-RANK applies the GVP-GNN framework to process and analyze 3D structures of protein complexes and score docking models by capturing geometric relationships within those structures.