Advancing Drug Repurposing with AI-Driven ATC Code Assignment

The Anatomical Therapeutic Chemical (ATC) classification system is vital for understanding a compound's therapeutic potential. However, manual assignment is often slow and complex. Researchers developed a novel multimodal sequence-to-sequence model to enhance ATC code prediction. This technology streamlines drug discovery and helps identify new uses for existing medications, which is particularly relevant for the pharmaceutical research landscape in India.

The Complexity of ATC Code Prediction

Assigning ATC codes involves navigating a hierarchical structure with four distinct levels. Traditional methods often struggle with polypharmacology, where a single drug exhibits multiple therapeutic behaviors. Furthermore, data scarcity for rare compounds makes accurate classification difficult. Consequently, a generative approach that leverages multiple data types provides a more robust solution for modern pharmaceutical research. Therefore, integrating diverse molecular representations is essential for future informatics.

How Multimodal Models Enhance Accuracy

The proposed model utilizes two distinct representations of chemical structures: SMILES codes and molecular descriptors. These modes provide complementary information. Specifically, SMILES represent the sequence-based structure, while descriptors capture physical properties. By integrating these modes, the model improves the precision of ATC code prediction. Moreover, a new decision method determines exactly when to stop generating labels, which prevents over-classification. Notably, this multimodal proposal clearly outperformed traditional baselines in comparative tests.

The study demonstrated superior performance across various benchmarks for both new and existing drugs. This breakthrough is particularly valuable for drug repurposing, where identifying hidden therapeutic traits can save years of clinical trials. In addition, the researchers made the source code publicly available. This allows Indian pharmaceutical scientists and academic researchers to adopt these automated tools into their discovery pipelines efficiently.

FAQ

What is the benefit of using SMILES and molecular descriptors together?

Using both SMILES and molecular descriptors allows the AI to see both the chemical sequence and the physical properties of a compound. This dual-mode approach provides a richer dataset, leading to higher accuracy in classifying complex drugs.

How does AI speed up drug repurposing?

AI identifies patterns in molecular structures that humans might miss. By predicting a drug's therapeutic class automatically, researchers can quickly see if an existing medication might work for a different disease without starting from scratch.

Disclaimer: This content is for informational and educational purposes only. It does not constitute medical advice or endorse any specific technology or drug. Refer to the latest local and national guidelines for clinical practice.

References

• Crozes T et al. A Multimodal Sequence-to-Sequence Model for Automatic Assignment of ATC Codes in Drug Discovery and Repurposing. J Chem Inf Model. 2026 Apr 14. doi: 10.1021/acs.jcim.6c00118. PMID: 41979032.


• Foster C. AI's Role In Accelerating Drug Discovery And Repurposing. Forbes. 2022 Dec 29.


• IBM Research. Tool-Augmented AI for Drug Discovery: An Agentic Platform for Integrated Molecule Generation and Screening. ACS Spring 2026.