Bayesian Integration of Tumor Mutational Signatures and Somatic Features Refines Pathogenicity Assessment of Germline Mismatch Repair Variants.

Bayesian integration of COSMIC tumor mutational signatures as an explicit likelihood ratio within ACMG/AMP germline classification

Paired germline and tumor sequencing data from 1,110 colorectal or endometrial tumors across 1,073 patients were analysed to quantify how mismatch repair-deficient mutational signatures can be integrated into germline variant interpretation. Using COSMIC single-base substitution signatures, the presence of an MMR-deficient signature increased the likelihood of an underlying pathogenic germline MMR variant roughly eightfold (LR ≈ 8; log10 LR ≈ 0.90), while its absence provided moderate-to-strong benign evidence (LR ≈ 0.156; log10 LR ≈ -0.81). Applied to 45 germline MMR variants of uncertain significance, joint modelling of tumor signatures with additional somatic and variant-level evidence produced clinically significant reclassification of 38 variants (84.4%): 3 to pathogenic or likely pathogenic and 35 to likely benign. Sixteen of the downgraded variants had been independently downgraded by Invitae.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Turning a tumor signature into an explicit likelihood ratio, and therefore into evidence usable within the Bayesian ACMG/AMP framework, is precisely what was missing: immunohistochemistry and MSI status are already routine but are still read alongside, rather than within, germline classification. Two limits: the likelihood ratio is derived from colorectal and endometrial cancers and does not transfer directly to rarer Lynch-spectrum tumors; and 35 of the 38 reclassifications move toward benign, which lightens the VUS burden without changing surveillance much. Directly implementable in any laboratory that already performs paired tumor sequencing.

Analysis by Dr Thibaut Benquey