CGINet

CGINet predicts chemical-gene interactions using a graph convolutional network applied to an integrated multi-relational graph of chemicals, genes, and pathways.

Key Features:

• Graph Convolutional Network (GCN) Framework: Uses graph convolutional networks to capture relationships within the integrated multi-relational biological graph for interaction prediction.
• Multi-Relational Graph Integration: Constructs a multi-relational graph with chemicals, genes, and pathways as distinct node types.
• Node Embedding Learning Perspectives: Supports whole-graph embedding learning and a subgraph approach that learns on binary association subgraphs (chemical-gene and gene-pathway) and transfers embeddings to a multi-interaction subgraph.
• Topological Structure Reconstruction: Reconstructs target-node topologies by modeling latent links within designed substructures.
• End-to-End Training: Trains encoder and decoder jointly using known chemical-gene interactions for end-to-end optimization.
• Model Variants: Includes CGINet-1, CGINet-2, and CGINet-3 implementations that incorporate different component combinations.

Scientific Applications:

• Drug Discovery: Identifies potential chemical-gene interactions to support discovery of new drug leads.
• Drug Repositioning: Predicts novel therapeutic targets for existing drugs to inform repositioning efforts.
• Biological Network Analysis: Provides insights into interaction patterns among chemicals, genes, and pathways within biological networks.

Methodology:

Applies graph convolutional networks on an integrated multi-relational graph of chemicals, genes, and pathways; learns node embeddings via whole-graph and subgraph perspectives (binary association subgraphs for chemical-gene and gene-pathway, transferred to a multi-interaction subgraph); reconstructs target-node topologies by modeling latent links in designed substructures; and jointly trains encoder and decoder using known chemical-gene interactions, with CGINet-1/2/3 representing different component configurations.