Harvard’s new AI model identifies potential treatments for rare diseases using existing drugs
Recent advancements in biological AI models have shown significant potential in the development of new medications. But can these technologies also be leveraged to find treatments for rare diseases using drugs that are already available? Researchers at Harvard Medical School aimed to explore this possibility with their new foundation model, TxGNN, which is specifically designed to identify existing drugs that could treat rare diseases lacking current therapies.
The TxGNN model, introduced in a paper published in Nature, is considered the first AI system to focus on repurposing already approved drugs to target rare diseases for which there are no treatments presently available on the market.
Although each rare disease individually affects a small number of people, the collective impact of the over 7,000 classified rare conditions is significant, affecting around 300 million people worldwide, according to The Lancet. Yet, only 5–7% of these conditions have drugs that are currently approved by the Food and Drug Administration (FDA). Interestingly, nearly one-third of all drugs approved by the FDA eventually gain approval for multiple uses, with some being applied to as many as 10 different indications, the authors of the study explain.
However, the challenge lies in the fact that discovering new applications for these drugs has historically been an unpredictable and serendipitous process. As the research team describes, it often relies on healthcare professionals accidentally uncovering these new uses while treating people.
“Predicting the efficacy of all drugs against all diseases would enable us to select medications with fewer side effects, design more effective treatments that target multiple points within a disease’s pathway, and systematically repurpose existing drugs for new therapeutic purposes,” the authors note.
The rise of AI-driven breakthroughs, such as Google’s AlphaFold protein prediction model, which won the Nobel Prize for Chemistry this year, has ignited an explosion of generative AI applications in drug discovery. Numerous companies are also developing models aimed at drug repurposing, according to GlobalData.
The Harvard team behind TxGNN claims their model is approximately 50% more effective than existing repurposing models in identifying potential drug candidates. Additionally, it is reported to be 35% better at predicting contraindications, which are reasons a specific drug may not be appropriate for an individual.
What sets TxGNN apart is its broader scope compared to many other AI models designed for drug discovery. As explained in a Harvard announcement of the research, while most models tend to concentrate on a single disease or a small group of related conditions, TxGNN is designed to comprehensively identify shared “mechanisms based on genomic underpinnings” between rare diseases and more common, better-understood conditions.
The tool has been trained on publicly available data sources such as DNA information and clinical notes and was validated using nearly 1.3 million de-identified patient records from Mount Sinai Hospital in New York. A notable feature of the TxGNN model is its explainer component, which clarifies the reasoning behind each prediction, offering a step-by-step breakdown of the decision-making process.
In a bid to encourage further scientific discoveries, the team has made TxGNN freely available to other researchers.
“This is precisely where we see the promise of AI in reducing the global disease burden, in finding new uses for existing drugs, which is also a faster and more cost-effective way to develop therapies than designing new drugs from scratch,” said Marinka Zitnik, one of the paper’s authors and an assistant professor of biomedical informatics at Harvard Medical School’s Blavatnik Institute.
This initiative highlights the potential of AI not just in the development of entirely new drugs but also in identifying existing medications that could be repurposed, offering faster, more efficient pathways to treatment for people living with rare diseases.