AI tool maps hidden links between diseases

An AI-powered tool from KAUST researchers is helping scientists trace hidden connections between diseases, revealing insights into how one illness might lead to another and, by extension, how treating one illness could help prevent another^[1].

By systematically combing through medical literature and real-world patient data, this tool maps cause-and-effect relationships, creating a framework that could guide targeted therapeutic strategies and uncover potential for drug repurposing.

Think of it as the ultimate disease relationship detective. Using natural language processing, the tool scans vast quantities of biomedical research to pinpoint causal connections — like how high blood pressure can set the stage for heart failure.

“Instead of treating diseases as unrelated outcomes, our approach facilitates the identification of shared risk factors among causally linked diseases,” says Sumyyah Toonsi, a graduate student in the Bio-Ontology Research Group. “This deepens our understanding of human diseases and enhances the performance of risk-prediction tools for personalized medicine.”

The tool’s power lies in its ability to go beyond mere association. Traditional methods might highlight which diseases commonly co-occur, but the KAUST tool — developed by Toonsi and her team under the guidance of computer scientist Robert Hoehndorf — identifies which diseases can trigger others.

For example, type 2 diabetes leads to high blood sugar, causing small blood vessel disease, ultimately resulting in a diabetic eye condition. Mapping these relationships suggests that treating one “upstream” condition may help prevent or lessen downstream complications.

To achieve these insights, the tool integrates scientific literature with data from the UK Biobank, a large-scale health database of about half a million Britons. This dual approach validates disease connections by checking that diseases follow a logical sequence, with causes preceding outcomes. This process strengthens the evidence of causation while highlighting new connections that might otherwise be overlooked.

Among its discoveries, the tool unearthed surprising links. As Toonsi explains, “We found endocrine, metabolic and nutritional diseases to be leading drivers of diseases in other categories,” including cardiovascular, nervous system and inflammatory diseases of the gut and eye. “This is interesting because many metabolic diseases can be managed with lifestyle changes, opening opportunities for broad disease prevention,” she says.

A standout feature is the tool’s ability to improve polygenic risk scores (PRS) — calculations that assess a person’s genetic susceptibility to disease. Standard PRS models don’t account for how one genetic variant might affect multiple diseases, but by adding causal disease relationships, the KAUST tool produces an enhanced PRS that improves prediction accuracy, especially for complex diseases.

This helps disentangle pleiotropic effects, where a single gene variant can impact multiple conditions. By factoring in these causal links, the tool offers a more holistic view of genetic risk.

Now freely available to the research community, this tool represents a major advancement for scientists exploring disease connections. Its potential applications range from refining prevention strategies to suggesting new uses for existing drugs. As researchers further investigate disease pathways, this tool could serve as a key resource in the quest to decode the interconnected landscape of human health.

Reference

1. Toonsi, S., Gauran, I.I., Ombao, H., Schofield, P.N. & Hoehndorf, R. Causal relationships between diseases mined from the literature improve the use of polygenic risk scores. Bioinformatics 40, btae639 (2024).| article.

---

---