Upgrading rigor and efficiency of germline cancer gene variant classification for the 2020s - ABSTRACT Since approximately 2010, the scale of clinical cancer predisposition genetic testing has increased dramatically. While a large fraction of sequence variants observed during testing are easily classified as benign or pathogenic, many others – principally missense substitutions, in-frame indels, and splice junction variants – are not easily placed on a spectrum from benign to clearly pathogenic. These are termed Variants of Uncertain Significance (VUS), and the clinical management of families in which they segregate would be improved if they could actually be classified. Consortium efforts to develop methods for evaluation and classification of VUS in BRCA1 and BRCA2 date back to a Breast Cancer Information Core satellite meeting held at the ASHG annual meeting in 2000; and methods that had been developing separately within the breast cancer, colorectal cancer, melanoma, and Li Fraumeni-syndrome genetics communities were cross-pollinated at a 2008 International Agency for Research on Cancer (IARC) working group meeting on VUS in cancer susceptibility genes. However, neither the qualitative nor the quantitative methods that sprouted from that meeting produced a generalizable overall approach. In 2015, the American College of Medical Genetics (ACMG) published guidelines for evaluating VUS across all Mendelian disease susceptibility genes. These guidelines produced a practical VUS evaluation framework that has been adopted by testing labs and organizations around the country. However, the ACMG system is entirely qualitative, with evidence weighted by expert opinion rather than by empirical evidence. Subsequently, we fitted the ACMG system into a quantitative Bayesian framework, providing a route to replacing qualitative evidence criteria from the ACMG system with empirically measured counterparts. Indeed, we hypothesize that there will be clear instances where strength accorded to current ACMG evidence criteria is contradicted by empirical measurement; correcting these will self-evidently improve the rigor of VUS evaluation. Aim 1 will place related ACMG data types into larger, logically consistent sets and then reduce or eliminate hidden dependencies between those sets. Noting that the ACMG variant classification guidelines were almost entirely qualitative, Aim 2 will improve the rigor of calibration for key data types through empirical measurement. Recently, we derived a quantitative Bayesian point-system for VUS evaluation and classification, which is back compatible with its parent quantitative Bayesian framework and the qualitative ACMG variant classification guidelines. Thus Aim 3 will refine this Bayesian point-system, taking advantage of the improved outputs from Aims 1 and 2. Finally, Aim 4 will benchmark elements of VUS evaluation and classification. Successful completion of these Aims will improve rigor in the system used for evaluation of VUS in cancer susceptibility genes, enabling higher throughput VUS evaluation and improving confidence in the resulting classifications.
