Clinical Evaluation of an AI System for Streamlined Variant Interpretation in Genetic Testing

The growing use of exome/genome sequencing to diagnose hereditary diseases has increased the interpretive workload for clinical laboratories. Efficient methods are needed to maximize diagnostic yield without overwhelming resources. We developed DiagAI, an AI-powered system trained on 2.5 million ClinVar variants to predict ACMG pathogenicity classes. DiagAI ranks variants, proposes diagnostic shortlists, and identifies probands likely to receive molecular diagnoses. It is powered by UP2, a machine learning–based score that predicts pathogenicity at the variant level, integrating molecular features, inheritance patterns, and phenotypic data when available. We retrospectively analyzed 966 exomes from a nephrology cohort, including 196 with causal variants and 770 undiagnosed cases. To benchmark UP2’s performance, we evaluated the ranking of 62 causal missense variants. UP2 ranked variants most effectively beyond shortlist sizes of 10 and identified pathogenic variants missed by AlphaMissense. While REVEL performed well for shortlists of 1–10 variants, it showed lower sensitivity than UP2 as list size increased. AlphaMissense and CADD were less effective overall. DiagAI identified 94.9% of causal variants in diagnostic exomes with HPO terms, compared to 90.8% without, with median shortlist sizes of 12 and 9 variants, respectively. It achieved a sensitivity of 57.1% and a specificity of 92.6% in tagging exomes likely to contain a diagnostic variant. With HPO terms, 74% of top-ranked variants were diagnostic, versus 42% without, and DiagAI outperformed Exomiser and AI-MARRVEL. DiagAI produces accurate shortlists that streamline variant interpretation and offers a scalable solution for growing diagnostic volumes without compromising quality.