Composite likelihood-based meta-analysis of breast cancer association studies

For detecting low risk disease variants in genome-wide association panels, meta-analysis is a powerful strategy to increase power. We apply a composite likelihood-based method, which models association with disease in regions defined on a linkage disequilibrium map and combines the evidence across multiple genome-wide samples. This fixed region approach has the advantage that, as only one statistical test is made per region, there is no increased multiple testing penalty in higher marker density panels. Imputation of missing genotypes is also advantageous to increase coverage. Meta-analysis of three breast cancer data sets combines evidence from samples that show heterogeneity in phenotype and, particularly, in marker coverage. The FGFR2 gene has the highest rank, consistent with previous analysis of one of these samples and supported by the small number of early-onset breast cancer cases included. The 8q24 breast cancer region also ranks highly and is supported by evidence from both early-onset and post-menopausal breast cancer samples. The PIK3AP1 gene region is highlighted in this analysis as a strong candidate for further study.