tested. Although perhaps appropriate where, for instance, all variants are predicted to have similar functional consequences due to directly disrupting or deleting a particular gene or due to affecting an intergenic region, this approach requires user intervention and is not universally applicable (e.g. in the context of a genome-wide analysis). A more general unsupervised approach for analysis of such complex loci, used in the studies examined here [10], [16] and also in analyses of common GSVs [36], [37], [39], is to test separately at multiple probe locations within a region. However, as illustrated by our analysis of the FOXP2 locus, it cannot be assumed that overlapping but distinct variants have similar effect sizes or even directions; furthermore, such ‘point-wise’ analysis entails multiple statistical tests at each GSV locus, leading to potential inflation in the reported P-value for the region as a whole. Thus, there is a need for methods that properly address the structural complexity frequently observed at GSV loci. One straightforward approach might be to apply a locus-specific multiple-testing correction, according to GSV complexity, to reflect the number of independent tests made at a locus, in a manner similar to that used to correct for multiple tests of SNPs
