Abstract Classifying germline variants in hereditary cancer genes remains challenging and requires integrating diverse lines of evidence. BoostDM is a computational method originally developed to identify somatic cancer driver mutations by detecting signals of positive selection. Given the functional overlap between somatic and germline pathogenic variants in cancer genes, we evaluated the utility of BoostDM for interpreting germline variants in hereditary cancer genes. We assessed BoostDM’s performance across six genes with dual roles in sporadic and hereditary cancer ( ATM , BRCA1 , BRCA2 , CDH1 , PTEN , TP53 ), using gene-specific BoostDM models. A total of 1275 germline single nucleotide variants with expert-reviewed pathogenic and benign classifications were included. BoostDM scores were compared to those from AlphaMissense and REVEL, two leading pathogenicity predictors for missense variants. BoostDM correctly classified 74.5% of pathogenic/likely pathogenic and 98.6% of non-synonymous benign/likely benign variants overall. It performed particularly well for non-synonymous, non-missense variants (92.3% sensitivity). For missense variants, BoostDM correctly identified 46% of pathogenic and 95.5% of benign variants. While BoostDM did not outperform AlphaMissense or REVEL, it demonstrated high specificity (99.5%) and positive predictive value (PPV = 98%) for missense variants with high scores ( > 0.5). Gene-specific performance varied, with TP53 showing the most robust results. In conclusion, BoostDM predictions are not a replacement for ACMG/AMP-guided germline variant classification, especially for missense changes. However, its high specificity and PPV suggest that high BoostDM scores can provide supportive evidence of pathogenicity, prompting further clinical and functional investigation.
Munté et al. (Mon,) studied this question.