GraphATC: advancing multilevel and multi-label anatomical therapeutic chemical classification via atom-level graph...

Area of Science:

• Pharmacology and Cheminformatics
• Drug Discovery and Development
• Computational Chemistry

Background:

• Accurate drug categorization within the Anatomical Therapeutic Chemical (ATC) system is crucial for drug development and research.
• Previous studies focused only on Level 1 ATC labels, overlooking the multilevel nature of drug classification.
• Existing benchmarks are outdated, lacking recent drug information integrated into the World Health Organization (WHO) ATC system.

Purpose of the Study:

• To address limitations in current drug categorization methods by developing a comprehensive, multilevel classification approach.
• To create a novel, updated benchmark dataset (ATC-GRAPH) encompassing all five ATC levels.
• To improve the representation and classification of complex drug types like Polymers, Macromolecules, and Multi-Component drugs.

Main Methods:

• Systematic cleansing and enhancement of drug datasets, integrating data from KEGG, PubChem, ChEMBL, ChemSpider, and ChemicalBook.
• Development of the ATC-GRAPH benchmark, expanding drug classification to include Level 2 and all subsequent levels.
• Application of graph-based learning techniques for precise molecular structure representation and classification.

Main Results:

• Established a novel benchmark, ATC-GRAPH, providing a comprehensive dataset for multilevel drug classification.
• Demonstrated improved accuracy in classifying Polymers, Macromolecules, and Multi-Component drugs.
• Achieved state-of-the-art performance in drug categorization tasks through extensive experimental validation.

Conclusions:

• The proposed graph-based learning framework offers a significant advancement in drug classification accuracy and fidelity.
• ATC-GRAPH serves as a valuable, up-to-date resource for researchers in drug development and pharmacology.
• Open accessibility of the benchmark, code, and web server promotes reproducibility and further research in the field.