Scientists at the Icahn School of Medicine at Mount Sinai have created a new artificial intelligence (AI) system that predicts whether people with rare genetic mutations are likely to develop disease. The method, detailed in the journal Science, aims to solve a longstanding challenge in genetic testing: interpreting the medical significance of unusual DNA changes.
The research team used AI models trained on more than one million electronic health records, drawing on routine laboratory results such as cholesterol, blood counts, and kidney function. By analyzing this everyday medical data, the system estimates a patient’s likelihood of developing conditions associated with specific genetic variants, providing a score from 0 to 1 that reflects disease risk.
Moving Beyond Black-and-White Results
Traditional genetic studies often categorize patients with a simple yes-or-no diagnosis. But many diseases, including cancer, diabetes, and heart disease, develop along a spectrum rather than fitting neatly into binary labels.
“Our goal was to move past the black-and-white answers that leave patients and providers uncertain,” said Ron Do, PhD, senior study author and Charles Bronfman Professor in Personalized Medicine at Mount Sinai. “By using AI and data that already exist in medical records, we can better estimate how likely disease will develop for someone with a particular variant. This approach supports precision medicine in a way that is both nuanced and scalable.”
The researchers generated what they call “machine learning penetration” scores for more than 1,600 genetic variants across 10 common diseases. A higher score suggested that a variant is more likely to contribute to disease, while a lower score pointed to minimal or no risk.
Unexpectedly, some variants previously labeled as “uncertain” showed strong disease signals, while others believed to cause illness appeared to have little impact.
Potential to Guide Medical Decisions
The study’s authors emphasized that the AI system is not a replacement for clinical judgment. Instead, it could be used as a guide for doctors facing unclear test results.
“For example, if a patient carries a rare variant associated with Lynch syndrome and the score is high, it might support earlier cancer screenings,” explained lead author Iain S. Forrest, MD, PhD. “On the other hand, if the score is low, unnecessary worry or overtreatment could be avoided.”
The team plans to expand the tool to cover more diseases, a broader range of genetic changes, and more diverse populations. Future research will also track whether people with high scores actually go on to develop disease and whether early interventions based on these predictions improve outcomes.
Toward a New Era of AI-Driven Genetics
The Mount Sinai group sees this work as part of a larger push to integrate AI into health care. The Windreich Department of AI and Human Health, which supported the research, is dedicated to applying artificial intelligence responsibly across clinical care, research, and education.
“Ultimately, our study points to a future where AI and everyday clinical data work together to provide clearer, more personalized insights for patients navigating genetic results,” said Do. “We want to give patients and families more confidence and help clinicians make better-informed decisions.”
The study, titled “Machine learning-based penetrance of genetic variants,” was supported by grants from the National Institutes of Health.
