Overview

• Researchers reported in Science on August 28 a machine-learning method that estimates variant penetrance for inherited conditions.
• Models trained on more than 1.3 million electronic health records generated continuous 0–1 risk scores for 1,648 rare variants across 31 genes and 10 autosomal-dominant diseases.
• The approach linked genotype to routine lab measures and found that some variants labeled uncertain showed disease signals, whereas some presumed pathogenic variants showed little effect.
• Authors say the scores are an adjunct to clinical judgment that could inform earlier screening or help avoid unnecessary interventions when risk appears low.
• Planned work includes expanding to additional diseases and variant types, testing in more diverse populations, and longitudinal follow-up to confirm predictive value.