Application of data science to inherited genetic variants of apoptosis genes and susceptibility to breast cancer

Background:
The identification of the breast and ovarian cancer genes BRCA1 and BRCA2 two decades ago has resulted in both improved patient risk assessment, and the use of PARP-inhibitors for treatment. In addition to highly penetrant mutations in genes like BRCA1 and BRCA2, hundreds of common inherited variants (single nucleotide polymorphisms or SNPs) have been identified through genome-wide association studies (GWAS), that also contribute to breast cancer susceptibility. However, identifying which SNPs are the functional risk variants, and elucidating their mechanism of action, remains a major challenge since most risk SNPs lie outside of gene coding regions. Thus although almost 200 breast-cancer associated SNPs have been identified, the functional basis has been determined for very few of these. We have shown that cancer-associated variants on chromosome 2q33 are related to local regulation of genes such as CASP8, CASP10 and CFLAR. These genes are involved in programmed cell death (apoptosis), which is highly dysregulated in cancer.

We hypothesize that inherited variants at apoptosis-related gene regions affect risk of breast cancer through mechanisms that modulate apoptosis.
Objectives:
1. To use SNP data for 15 apoptosis gene regions to fully characterise the associations with breast cancer in each region (genetic fine-mapping).
2. To identify the credible functional variants and their target genes in each region, and prioritise SNP-gene pairs for future laboratory-based functional studies.
3. To examine statistical interactions between the gene regions and map these to functional networks to yield a pathway-wide mechanistic view of susceptibility.

Experimental approach:
1. We have SNP data for 15 apoptosis gene regions from the Breast Cancer Association Consortium (BCAC), comprising over 120,000 cases and 105,000 controls. The results from standard univariate and multivariate stepwise and penalised logistic regression approaches
will be compared to those using novel Bayesian techniques that incorporate prior functional information.
2. SNPs identified in (1) will be intersected with publicly available expression QTL and ENCODE epigenetic data, and collaborative RNAseq and 3D chromatin conformation data.
3. Logistic framework and machine learning techniques will be used to examine interactions, both pairwise and higher-order.