Accurate interpretation of genetic variants still represents a major challenge. According to current recommendations from the American College of Medical Genetics and Genomics (ACMG), variant interpretation relies on a comprehensive analysis including, among others, computational data for prediction of variant pathogenicity. However, the predictive accuracy of in silico tools is often limited, and results are frequently inconsistent. In the current study, we evaluated the predictive performance of a previously described innovative classifier (MutScore) for missense variants in our cohort of probands with inherited cardiac diseases (InCDs).
We retrospectively reviewed missense variants detected in our cohort of probands with InCDs. Variants were analyzed with four in silico tools commonly used in our diagnostic pipelines (CADD, Polyphen-2, Alpha-missense and Revel) and with MutScore, a new meta-predictor combining data on variant location with the output of 16 existing predictors. For each variant, we recorded the original classification (established according to scientific evidence available at the time of molecular diagnosis) and the updated classification performed at the present time, according to ACMG standards.
We detected 252 missense variants in our cohort of 517 patients affected by InCDs. MutScore was the most proficient tool in classifying variants (0.89 maximum area under the curve [95% confidence interval (CI) 0.85-0.94]). Compared to Revel, the second-best predictor, MutScore showed superior sensitivity (73% vs 57%) at the maximum tolerated false-positive rate of 10%, higher specificity (0.83 vs 0.36) and a markedly lower false-positive rate (0.17 vs 0.64), supporting a more nuanced and accurate assessment, especially for benign or likely benign variants. MutScore also appeared to perform better for variants located in genes associated with channelopathies than for variants in cardiomyopathy-related genes. Notably, when comparing the original and updated classification, 27% (69/252) of missense variants underwent a change in classification over the 9-year follow-up period. Among these, reclassification had a significant impact on clinical management in one third of cases (i.e., variants of uncertain significance upgraded to pathogenic or likely pathogenic variants or vice versa), with a 4.8% increase in molecular diagnosis of InCDs over the 9-year period.
Our study supports the excellent performance of MutScore in a real-life dataset of missense variants associated with the rare subset of InCDs. MutScore represents a promising application of artificial intelligence with major potential in cardiogenetics to improve diagnostic precision in clinical practice. In addition, our results highlight the importance of periodic reanalysis of variants, incorporating newly available scientific evidence, as attested by the significant implications for patient management and clinical decision-making.