Rare and undiagnosed diseases affect more than 300 million people worldwide, representing an immense human and economic burden. These conditions, although rare individually, collectively represent an enormous health challenge. Despite the urgency, only 5 to 7 percent of these diseases have an FDA-approved drug, leaving many patients without effective treatment options.
A revolutionary artificial intelligence (AI) tool offers a new path to fill this gap. Developed by researchers at Harvard Medical School and published in the journal, Natural medicinethe AI model, called TxGNN, has demonstrated unprecedented potential for discovering therapies for rare and neglected diseases.
By repurposing existing drugs, TxGNN has identified therapeutic candidates for more than 17,000 diseases, including those for which no treatments are currently available. This achievement marks a significant step forward in drug discovery.
The AI revolution in drug reuse
TxGNN is a graph-based AI model specifically designed for drug reuse. Unlike traditional approaches that narrowly focus on diseases with existing therapies, this tool analyzes features common to all diseases to identify new drug candidates. It uses a large data set, including genomic information, cell signaling data and clinical records, to establish links between diseases and potential treatments.
“With this tool, we aim to identify new treatments across the entire disease spectrum,” said Marinka Zitnik, principal investigator and assistant professor of biomedical informatics at Harvard Medical School. “For rare and neglected diseases, this model could help close a gap that creates serious health disparities. »
In its initial trials, TxGNN screened nearly 8,000 investigational and FDA-approved drugs, identifying potential uses for more than 17,000 conditions. Its forecasts also included information on possible side effects and contraindications, providing a level of transparency that builds physician confidence.
Expanding the scope of drug reuse
Repurposing existing drugs provides a faster and more cost-effective route to therapy development. Many drugs have effects beyond their original targets, some of which were not discovered in early clinical trials. Over time, nearly 30 percent of drugs approved by the FDA have acquired additional indications, with some gaining many new uses.
Traditionally, drug reuse has relied on incidental discoveries or off-label uses guided by clinician intuition. TxGNN aims to replace this serendipitous approach with a strategic, data-driven methodology. The model’s ability to identify genomic underpinnings common to all diseases allows it to extrapolate from well-studied conditions to those that are poorly understood.
“We have tended to rely on luck and chance rather than strategy, which limits drug discovery to diseases for which drugs already exist,” Zitnik explained. “Even for common illnesses, new drugs could offer alternatives with fewer side effects or better effectiveness for some patients. »
Superior performance and transparency
TxGNN outperformed leading AI models for drug repurposing, demonstrating nearly 50% better accuracy in identifying drug candidates and 35% better prediction of contraindications. This performance was validated using 1.2 million patient records and various experimental tasks.
For example, the model successfully identified potential treatments for three rare diseases that it had not encountered during training: a neurodevelopmental disorder, a connective tissue disease, and a genetic disease affecting fluid balance.
What sets TxGNN apart is its “Explainer” feature, which provides detailed rationales for its predictions. This transparency brings the model closer to human clinical reasoning, providing multi-hop medical knowledge pathways for clinicians to review. “This ability to reason and explain builds confidence in the model’s recommendations,” Zitnik noted.
Additionally, the tool’s predictions often aligned with existing off-label uses, thus validating its reliability. For example, in one task, TxGNN identified small molecules capable of targeting proteins involved in pathogenic pathways. These results not only correspond to current medical knowledge, but also highlight the potential of the model to discover new therapeutic avenues.
Closing treatment gaps
The implications of TxGNN extend far beyond rare diseases. For conditions with existing treatments, the model could identify alternatives with fewer side effects or improved effectiveness. By reusing drugs with known safety profiles, TxGNN accelerates the clinical application process, avoiding the time-consuming and costly process of developing new drugs from scratch.
The research team has made the tool freely available, encouraging clinician-scientists to explore its capabilities. Several rare disease foundations have already partnered with the team to identify potential treatments for conditions with limited or no treatment options.
While therapies identified by TxGNN require further validation, the model’s unprecedented capability for drug reuse represents a transformative step in medicine. “This is precisely where we see the promise of AI in reducing the global burden of disease,” Zitnik said. “Finding new uses for existing drugs is faster, more cost-effective and can potentially save countless lives. »
