FACT: Feature Aggregation and Convolution with Transformers for predicting drug classification code.

Area of Science:

• Pharmacology
• Computational Chemistry
• Bioinformatics

Background:

• Drug repositioning accelerates drug development by finding new uses for existing drugs.
• Anatomical Therapeutic Chemical (ATC) codes provide a systematic framework for drug classification and prediction.
• Existing ATC prediction methods face challenges due to the complex ATC hierarchy and scalability issues.

Purpose of the Study:

• To develop a novel and accurate method for predicting Anatomical Therapeutic Chemical (ATC) codes for drug repositioning.
• To address the limitations of existing ATC prediction approaches, particularly at higher hierarchical levels.

Main Methods:

• Proposed a new approach named Feature Aggregation and Convolution with Transformer models (FACT).
• Computed three types of drug similarities, incorporating ATC code similarity with hierarchical weights and masked drug-ATC code associations.
• Utilized a convolution-transformer encoder to generate embeddings for predicting drug-ATC code associations.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.9805 and an Area Under the Precision-Recall Curve (AUPRC) of 0.9770 at ATC level 4.
• Outperformed previous methods by 15.05% in AUROC and 18.42% in AUPRC.
• Demonstrated the effectiveness of integrating diverse drug features and transformer models for ATC code prediction.

Conclusions:

• The FACT model significantly enhances the accuracy of ATC code prediction.
• The study highlights the potential of transformer-based models in drug repositioning and pharmaceutical research.
• The developed method offers a scalable and effective solution for navigating the complexities of the ATC classification system.