AI Model TxGNN Repositions Drugs for Rare Diseases

Of the more than 7,000 rare diseases worldwide, only about 5-7% have corresponding FDA-approved drug treatments. A team of researchers from Harvard Medical School and their collaborators have developed an AI model called TxGNN, the first new approach developed specifically to identify drug candidates for rare diseases and drug-free conditions. The findings were published in the journal Nature Medicine.

About TxGNN

Drug repurposing identifies new therapeutic uses based on safety and efficacy data from existing drugs. Drug repurposing can significantly improve the efficiency of clinical applications and reduce development costs compared to de novo development.

TxGNN is a fundamental AI model based on graph neural networks (GNNs) dedicated to zero-shot drug repurposing. TxGNN embeds complex relationships between diseases and drugs into a latent representation space optimized to reflect the relationships between them by training medical knowledge graphs (KGs). The KGs contain vast medical concepts such as 17,080 diseases and 7,957 drugs, which provides a rich data base for the training of TxGNN.

The researchers also developed the TxGNN interpreter module to help physicians and researchers understand the model's prediction logic. This module demonstrates the potential connection between drugs and diseases through multi-hop paths, which greatly improves the interpretability and trustworthiness of the model.

Fig. 1. The TxGNN model was trained on complex medical KGs. (Huang K.; et al. 2024)

TxGNN Performance

• TxGNN identifies drug candidates for more than 17,000 diseases from existing drugs.
• TxGNN improved the accuracy of its indications and contraindications predictions by 49.2% and 35.1%, respectively, compared to eight existing methods in a rigorous zero-sample evaluation.
• Many of TxGNN's predictions were highly consistent with actual clinical drug use. For example, in predicting potential therapeutic agents for Wilson's disease, TxGNN recommended Deferasirox as the most promising candidate. This drug is already used in the clinic for the treatment of iron overload disorders, and TxGNN demonstrated through its explanatory module that Deferasirox may have a positive effect on the treatment of Wilson's disease through metabolic pathways. This prediction is also consistent with relevant findings in the medical literature, demonstrating that the model is scientifically sound.
• TxGNN is not only capable of finding potential novel uses in existing therapeutic regimens, but it also can accurately predict possible drugs in the absence of known therapeutic regimens.

TxGNN offers a method for drug repurposing in the face of stale and expensive development cycles, and could fill the cure gap for rare diseases. The performance of TxGNN shows that multi-disease predictive modelling is viable as a strategy for drug repurposing, particularly for diseases lacking sufficient data.

As AI advances, AI technology will continue to become a more important part of drug discovery and personalized medicine. We unite the most powerful AI technologies and are dedicated to offering clients the most advanced AI models in drug development in an effort to speed up drug development cycles and costs, and get effective treatments to patients faster. Get in touch with us if you are interested in our services to know more about how our AI can drive your business to grow.

Related Services

Drug Repositioning and Repurposing
AI-powered Drug Discovery and Design
AI-assisted Drug High Throughput Screening