DACPGTN

DACPGTN predicts Anatomical Therapeutic Chemical (ATC) codes for drugs by integrating drug, disease, and target information with graph-based deep learning to support drug classification, repositioning, and toxicity inference.

Key Features:

• Composite Feature Construction: Integrates diverse biomedical information to construct composite features for drugs, diseases, and targets.
• Graph Transformer Network Inspiration: Leverages concepts from Graph Transformer Networks to learn potential novel interactions among drugs, diseases, and targets.
• Heterogeneous Network Utilization: Builds drug–target–disease heterogeneous networks that encapsulate comprehensive interaction information for ATC prediction.
• Graph Convolutional Networks (GCN): Employs Graph Convolutional Networks to generate embeddings for drug nodes from composite features and network topology.
• Multi-Label Learning: Performs multi-label classification to predict multiple applicable ATC codes per drug.

Scientific Applications:

• Drug screening: Supports classification and prioritization of compounds via predicted ATC labels.
• Drug repositioning: Identifies potential new therapeutic uses for existing drugs by revealing alternative ATC classifications.
• Similarity research: Facilitates similarity analyses using embeddings derived from heterogeneous drug–disease–target networks.
• Indication and toxicity inference: Provides insights into potential indications and toxicities through ATC code predictions.
• Therapeutic development support: Informs decision-making in drug development by supplying ATC-based classification and safety-related signals.

Methodology:

Constructs a drug–disease–target heterogeneous network and composite features, applies Graph Transformer Network-inspired learning and Graph Convolutional Networks to generate drug embeddings, and uses multi-label classification to predict ATC codes.