Researchers at the Icahn School of Medicine at Mount Sinai have developed a new approach that combines artificial intelligence with routine laboratory tests to predict the likelihood of disease development from rare genetic variants. This innovation addresses a major challenge in genetics: interpreting the real-world impact of rare DNA mutations, which often come with uncertain clinical significance. Traditional genetic testing frequently leaves patients and doctors unsure about whether a rare variant will lead to disease. To solve this, the team created machine learning models that analyze vast amounts of real-world health data, including common lab results such as cholesterol levels, blood counts, and kidney function. These models use electronic health records to assess disease risk on a spectrum rather than a simple yes-or-no basis, offering a more accurate and personalized picture of genetic risk. The study, published in Science under the title "Machine learning-based penetrance of genetic variants," used data from over 1 million patient records to develop AI models for 10 common diseases, including heart disease, diabetes, and certain cancers. The researchers then applied these models to individuals with known rare genetic variants, generating a risk score between 0 and 1. A score closer to 1 indicates a higher likelihood of disease, while a lower score suggests minimal risk. The findings revealed surprising insights. Some variants previously classified as "uncertain" showed strong disease associations, while others thought to be harmful had little impact in real-world data. This highlights the limitations of relying solely on genetic databases or theoretical models. Lead author Iain S. Forrest, MD, PhD, explained that the AI-generated "ML penetrance" scores could help guide clinical decisions. For example, a high-risk score for a Lynch syndrome variant might prompt earlier cancer screenings, while a low score could prevent unnecessary anxiety or invasive interventions. The researchers emphasize that the AI tool is not meant to replace medical judgment but to support it—especially when test results are ambiguous. The team is now expanding the model to include more diseases, a broader range of genetic changes, and more diverse populations. They also plan to track long-term outcomes to validate whether high-risk predictions translate into actual disease development and whether early action improves health outcomes. Ron Do, PhD, senior author and Charles Bronfman Professor in Personalized Medicine, said the work represents a step toward a future where AI and everyday clinical data work together to deliver clearer, more actionable genetic insights. The goal is to empower patients and providers with greater confidence and precision when interpreting genetic test results, advancing the promise of personalized medicine.