Integration of Machine Learning-Based Pathogenicity Prediction and Phenotype Matching Improves Variant Prioritization in Rare Clinical Testing
Every exome returns thousands of variants. Somewhere in that list is the one causing a patient's disease, and finding it is still the slowest step in clinical genomics.
A new peer-reviewed study in Current Issues in Molecular Biology puts hard numbers on how well DiagAI solves that problem. Across 196 diagnosed exomes from adults with kidney disease of unknown cause:
- DiagAI's shortlist captured the causal variant in 94.9% of cases when patient symptoms were provided, and 90.8% without them.
- The single top-ranked candidate was the correct diagnosis 74% of the time with symptoms, versus 42% without, outperforming Exomiser and AI-MARRVEL on the same cohort.
- The underlying pathogenicity model, UP2, placed causal variants within the top 100 candidates in 87% of cases, well ahead of REVEL's 61%.
The result is a shortlist of about ten variants that consistently contains the answer, cutting the manual review workload for clinical scientists as sequencing volumes keep growing.
Read the full study to see how DiagAI combines a machine-learning pathogenicity score, phenotype matching, and expert rules on inheritance and sequencing quality into one ranked shortlist.
