By Deborah Borfitz 

November 20, 2024 | Harvard scientists, led by Dr. Marinka Zitnik, have developed an AI-based tool, TxGNN, to identify drug repurposing opportunities for diseases lacking effective treatments. The AI model harnesses a novel approach by focusing on multiple diseases simultaneously, allowing insights from well-documented conditions to be applied to rare diseases with sparse data. Traditionally, drug-repurposing models concentrate on individual diseases, limiting broader applicability. TxGNN, however, employs a “zero-shot” framework, predicting therapeutic candidates for diseases with limited or no known treatments among the 7,000 rare diseases affecting populations worldwide. 

TxGNN’s methodology centers around U.S. FDA-approved drugs and aims to target disease-related networks, identifying direct or indirect therapeutic effects. The model consists of two primary modules: one suggesting drug indications and contraindications, and another that explains the knowledge graphs linking drugs to diseases. In a recent Nature Medicine study, TxGNN ranked about 8,000 drugs, assessing treatment potential for over 17,000 diseases. Its top predictions frequently aligned with existing off-label drug prescriptions, outperforming other drug-repurposing AI models. 

Importantly, TxGNN’s functionality extends beyond prediction; its “Explainer” module provides insights into how medical knowledge influences the AI’s recommendations. This is particularly useful for researchers evaluating drug efficacy and patient safety. In real-world trials, the Harvard team observed that this feature raised confidence levels among clinicians and researchers assessing TxGNN’s predictions. 

Further developments are underway, supported by $48.3 million in funding from the Advanced Research Projects Agency for Health. Partnering with BioPhy, TxGNN’s results are being tested in clinical trials, with Every Cure—a nonprofit committed to advancing drug repurposing efforts—playing a key role in deploying these tools for high-value drug-disease matches. 

TxGNN is currently available to the public, enabling researchers and clinicians to input disease and drug queries for immediate AI-driven predictions. The tool aims to democratize access to AI resources in drug development, addressing the critical needs of underserved patient groups with rare diseases. Harvard’s collaboration with patient organizations, including the Chan Zuckerberg Initiative, highlights the social impact of TxGNN by supporting foundations dedicated to advancing treatments for rare conditions. 

Read more about the work at Bio-IT World.