Stability selection for genome-wide association.

This article applies the recently proposed "stability selection" procedure of Meinshausen and Bühlmann to the problem of variable selection in genome-wide association. In particular, it explores whether stability selection can identify new regions of interest originally missed or can call into legitimate question regions originally flagged. Our analysis of the seven data sets of the Wellcome Trust Case-Control Consortium suggests that stability selection effectively controls the family-wise error rate but suffers a loss of power. The extensive correlation structure among SNP markers induced by linkage disequilibrium renders the procedure too conservative, causing it to miss regions known to be highly significant from simple marginal analyses. As a remedy one can aggregate nearby SNPs into groups and select groups rather than individual SNPs. The modified procedure can accurately identify the most important regions of genome-wide association, but in a simulation study it still offers less power than simpler and less computationally intensive methods of marginal association testing.