A0416
Title: Large scale case-control analyses using LR modelling improves rare variant classification in breast cancer risk genes
Authors: Damianos Michaelides - The Cyprus Institute of Neurology and Genetics (Cyprus) [presenting]
Abstract: A large number of rare variants in breast cancer susceptibility genes remain as variants of uncertain significance (VUS). The impact of hundreds of VUS is quantified through evidence derived from analysis of over 15K variants in PALB2, TP53, CHEK2, and ATM. The statistical methodology employed benefits from the largest breast cancer case-control dataset to date (>60K cases, >250K controls), using sequencing data. The ccLR method is used, which models the likelihood of observing a variant in cases versus controls, incorporating survival information and known gene-level penetrance. However, the method assumes a uniform risk across all variants in a gene, ignoring that individual variants may confer stronger or weaker risks, hence, limiting its ability to detect risk heterogeneity. This is addressed by a refined approach using a dynamic scaling parameter that calibrates relative risk per-variant. This allows the model to evaluate how different risk magnitudes fit the data for each variant. The refined method is useful in identifying variants that carry a higher or lower risk than the gene's average pathogenic variants. LRs were computed across a range of scaling values, representing a spectrum of relative risks. Pathogenic evidence was based on the maximum LR. Benign evidence was based on the LR at the gene-level risk. The refined approach provides evidence for 368 rare unclassified variants, a 15\% increase over the evidence identified by the standard method.
