---

Artificial intelligence (AI) continues to transform the healthcare landscape, offering innovative solutions to some of the most pressing medical challenges. One of the most promising developments is the use of AI to identify drug repurposing opportunities for diseases without effective treatments. Harvard Medical School’s innovative model, TxGNN, represents a groundbreaking step in this direction.

Introducing TxGNN: A New Approach to Drug Repurposing

Developed by Dr. Marinka Zitnik and her team, TxGNN is a graph foundation model designed to predict therapeutic candidates for diseases with sparse or no treatment options. Unlike traditional drug-repurposing models that focus on a single disease, TxGNN is trained across a wide array of conditions. This enables the model to transfer insights from diseases with abundant molecular data to those with limited data, addressing the unmet needs of thousands of rare diseases.

Zero-Shot Predictions for Rare Diseases

TxGNN operates as a “zero-shot” model, capable of predicting drug-disease pairings even in the absence of direct training data for those conditions. With over 7,000 rare diseases worldwide, most of which lack effective treatments, this capability is transformative. The model focuses on FDA-approved drugs, significantly reducing the time and cost of bringing therapies to market.

The core principle behind TxGNN is leveraging disease-perturbed networks to identify drugs that either directly target these networks or propagate therapeutic effects indirectly. The model includes two key modules:

1. Prediction Module: Identifies potential drug indications and contraindications.
2. Explainer Module: Provides insights into the knowledge graphs used to generate predictions, enhancing transparency and usability.

Performance and Real-World Validation

A recent study published in Nature Medicine demonstrated TxGNN’s ability to rank nearly 8,000 FDA-approved and clinical trial drugs across more than 17,000 diseases. Remarkably, 92% of the diseases lacked known molecular target interactions. The model outperformed eight other AI drug-repurposing tools, offering more accurate predictions and fewer contraindications.

Many of TxGNN’s top-ranked drug-disease pairs aligned with off-label prescribing practices, suggesting real-world relevance. Additionally, the model has been validated through collaborations with institutions like Mount Sinai, where curated electronic medical records (EMRs) were used to refine predictions.

AI for Public Good

Dr. Zitnik emphasizes that TxGNN represents “AI for public good,” targeting underserved diseases with small patient populations. The tool is freely available to researchers, clinicians, and patient advocacy groups. Users can explore the model’s predictions through an interactive website and even train the model on private datasets to suit their needs.

Partnerships and Future Directions

The impact of TxGNN extends beyond academia. Collaborations with organizations like the Chan Zuckerberg Initiative and Every Cure aim to evaluate high-value drug-disease matches and optimize clinical trial designs. The model’s success has also attracted interest from major pharmaceutical companies seeking to integrate it into their workflows.

Looking ahead, Dr. Zitnik’s team plans to enhance TxGNN by integrating patient-level data and expanding its scope to include infectious diseases. This involves creating more comprehensive knowledge graphs that account for host-pathogen interactions and leveraging real-world treatment data to improve prediction accuracy.

Transforming Drug Development

TxGNN represents a paradigm shift in drug repurposing, offering a scalable, cost-effective solution for diseases that have long been neglected. By bridging the gap between data-rich and data-scarce conditions, AI tools like TxGNN have the potential to bring new hope to patients worldwide.

Innovator Spotlight

Dr. Marinka Zitnik is a pioneering assistant professor of biomedical informatics at Harvard Medical School, recognized for her groundbreaking work in artificial intelligence and its applications in healthcare. She is an associate faculty member at the Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University. Dr. Zitnik’s research focuses on developing advanced AI models that tackle complex challenges in drug discovery, precision medicine, and disease diagnosis. Her notable contributions include the creation of TxGNN, a zero-shot drug-repurposing model that leverages knowledge graphs to identify treatment opportunities for rare and underserved diseases. 

---