DACPGTN: Drug ATC Code Prediction Method Based on Graph Transformer Network for Drug Discovery.

Area of Science:

• Pharmacology
• Computational Biology
• Drug Discovery

Background:

• The Anatomical Therapeutic Chemical (ATC) classification system is crucial for drug screening, repositioning, and similarity research, aiding in understanding drug indications and toxicity.
• Accurate ATC code prediction accelerates drug development and therapeutic application.

Purpose of the Study:

• To propose an end-to-end model, DACPGTN, for predicting drug ATC codes.
• To enhance drug discovery through accurate ATC code prediction.

Main Methods:

• DACPGTN constructs composite drug, disease, and target features using diverse biomedical information.
• It employs a Graph Transformer Network to learn novel interactions, creating drug-target-disease heterogeneous networks.
• Graph convolution networks generate drug node embeddings for multi-label learning tasks.

Main Results:

• The DACPGTN model demonstrated superior prediction performance compared to existing methods on benchmark datasets.
• The model effectively utilizes comprehensive interaction information within heterogeneous networks.

Conclusions:

• DACPGTN offers a powerful approach for predicting drug ATC codes, facilitating drug discovery and development.
• The model's ability to learn from heterogeneous networks highlights its potential in pharmaceutical research.