How an AI tool could find new treatments for rare diseases

Currently, there are over 7,000 rare and undiagnosed diseases worldwide, affecting around 300 million people. However, only 5% to 7% of these conditions have any available treatments, which means many are left untreated or undertreated.

To address this issue, researchers at Harvard Medical School developed a new AI model called TxGNN. The model is the first tool developed specifically to identify potential treatments for rare diseases and conditions. It was trained on vast amounts of data, including DNA information, cell signaling, levels of gene activity, and clinical notes.

The tool has two main features: one identifies treatment candidates and potential side effects while the other explains the reasoning behind the decision. According to the researchers, the second feature provides transparency to increase physicians' confidence in the tool.

So far, the tool has identified drug candidates from almost 8,000 medications for 17,080 rare diseases, including conditions that don't have any treatments available. The potential drug candidates include both FDA-approved medications and experimental drugs currently being tested in clinical trials.

In a study published in Nature Medicine, the researchers found that TxGNN was on average almost 50% better at identifying treatment candidates when compared to leading AI models used for drug repurposing. In addition, TxGNN was 35% more accurate at predicting which drugs would have contraindications.

However, the researchers acknowledged that the tool's success is only as good as the medical knowledge that it uses to make its assessments. "Challenges such as data biases and potentially outdated information in the knowledge graph must be addressed," they wrote.

Commentary

When it comes to repurposing drugs, Marinka Zitnik, an assistant professor of biomedical informatics at Harvard Medical School's Blavatnik Institute and the model's lead researcher, said that people have "tended to rely on luck and serendipity rather than on strategy, which limits drug discovery to diseases for which drugs already exist."

According to the researchers, TxGNN is better than most current AI models used for drug discoveries since it was trained in a way that allows it to use existing data to make new predictions. This ability means the reasoning the tool uses is closer to the reasoning a human clinician might use to generate new ideas if they had access to all the data the AI uses.

Although the researchers cautioned that any treatments identified by the model would need further evaluation for dosing and timing of delivery, they said it would help speed up drug repurposing in a new way.

Currently, researchers have made the AI model available for free to encourage further research on potential treatments for rare diseases. The team is also collaborating with several rare disease foundations to potentially identify new treatments.

"With this tool we aim to identify new therapies across the disease spectrum but when it comes to rare, ultrarare, and neglected conditions, we foresee this model could help close, or at least narrow, a gap that creates serious health disparities," Zitnik said.

"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," Zitnik added.

Advisory Board's AI resources

For more insights in AI and healthcare, check out these Advisory Board resources:

This expert insight outlines different strategies for AI adoption, which can help you decide the best approach for your organization. Similarly, this expert insight explains how health system executives currently approach AI in healthcare.

To use AI effectively, this infographic explains how to overcome AI challenges to unlock the technology's full potential while this field guide outlines how to take a problem-first approach to AI.

You can also read our research on the use of AI in different areas of healthcare, including cardiovascular care and imaging.

We also have a featured page on AI, which includes research on how to mitigate challenges with the technology, how leaders should approach investments in AI tools, and more.  (Murphy, Becker's Hospital Review, 9/25; Reed, Axios, 9/25; Pesheva, The Harvard Gazette, 9/25; Huang et al., Nature Medicine, 9/25)