GNN-SubNet

GNN-SubNet identifies disease-associated subnetworks within protein-protein interaction (PPI) networks by applying explainable graph neural networks that integrate multimodal molecular data to link subnetworks to disease mechanisms.

Key Features:

• Implementation: Python-based graph deep learning framework that operates on PPI networks.
• Graph-Based Framework: Employs a novel modification of the GNNexplainer to produce model-wide explanations and detect potential disease subnetworks by analyzing PPI topology per patient.
• Multimodal Node Features: Nodes are enriched with molecular data including gene expression and DNA methylation profiles.
• Explainability: Uses an advanced GNNexplainer variant to provide interpretable links between specific subnetworks and disease associations.
• Practical Applications: Applied for drug target recognition within protein-drug interaction networks and for cancer gene discovery.

Scientific Applications:

• Drug target recognition: Identification of subnetworks relevant to drug–protein interactions for target discovery.
• Cancer gene discovery: Detection of disease-specific subnetworks to prioritize cancer-associated genes.
• Understanding disease mechanisms: Linking multimodal molecular alterations in PPI subnetworks to disease biology and personalized medicine research.

Methodology:

Operates a graph-based deep learning pipeline in Python that inputs patient-specific PPI topologies with nodes annotated by gene expression and DNA methylation, and applies a modified GNNexplainer to obtain model-wide, interpretable subnetworks.