Improving the Power of Rare-Variant Association Tests by a Position-Dependent Kernel Approach

Current high-throughput sequencing technologies have allowed for an easy and cost-efficient identification of rare single-nucleotide variations (SNVs), many of which have already been proven to be associated with diseases or complex traits. Despite these successes, genomewide association studies involving rare variants remain statistically challenging. Classical single-SNV association studies particularly suffer from poor statistical power, as the potentially large number of SNVs often leads to poor significance upon false discovery rate (FDR) correction. To overcome these difficulties, approaches have been proposed that do not consider all SNVs individually; instead, they group SNVs and perform tests on those groups. This can either be done by grouping SNVs that are in the same genomic region of interest (e.g. the same transcript or exon) or by windowing along each chromosome. The choice of the groups/windows is crucial: FDR correction does not pose a serious problem if there are large, and consequently fewer, windows, but the local tests have poor power for large windows. If smaller windows, and consequently a larger number thereof, are chosen, the local tests perform well, but FDR correction nullifies this advantage. For more see http://www.bioinf.jku.at/publications/2012/HGV2012_Bodenhofer.pdf