Researchers at the Icahn School of Medicine at Mount Sinai have developed a new artificial intelligence-based method to evaluate the association between rare genetic mutations and diseases. The method uses machine learning to analyze routine laboratory test data, providing a quantitative assessment of the penetrance of genetic variants. The research results were published in the journal Science.
Traditional genetic studies often use binary diagnostic classifications, but many diseases exhibit continuous spectrum characteristics, leading to frequent uncertainty regarding the clinical significance of rare genetic mutations. The research team utilized more than one million electronic health records to build AI models for 10 common diseases. Through routine test indicators such as cholesterol levels and blood cell counts, they quantified disease risk scores for carriers of genetic variants.
Senior author of the study Dr. Ron Do stated: "We hope to move beyond black-and-white answers, because this pattern often makes it difficult for patients and healthcare providers to understand the actual meaning of genetic test results." The new method generates an "ML penetrance" score ranging from 0 to 1, with higher scores indicating a greater likelihood that the variant will cause disease.
The research team evaluated more than 1,600 genetic variants. Some variants previously labeled as "of uncertain significance" showed clear disease signals, while certain variants considered pathogenic had minimal impact in real-world data. Lead author Dr. Iain S. Forrest pointed out: "Although AI models cannot replace clinical judgment, they can provide important references for test results with unclear interpretations, helping to make decisions on early screening or preventive measures."
This method is expected to help doctors develop personalized management plans based on the scoring results. For carriers of high-risk variants, early screening is recommended, while low-risk variants can avoid unnecessary interventions. The research team plans to further expand the model's coverage of diseases, increase the variety of genetic variants and population diversity, and conduct long-term tracking of prediction accuracy.
