be avoided. Even after the use of predefined criteria to select a subset of GSVs for analysis, the necessary correction remains substantial: In the previous studies, subjects with common obesity carried approximately 20 GSV calls per individual (1080 cases, 2500 controls) [16]; more than 300 separate rare GSV loci were identified in the extreme obesity discovery cohort alone [10]; and both used a point-wise method for assessing association at complex GSV regions. Although attempts were made to take account of multiple testing (e.g. exclude GSVs present in the Database of Genomic Variants, including only rare GSVs, empirical estimate of FDR), it is unclear to what extent these methods are effective – methods developed for association analysis of rare sequence variants are not consistently well-powered even for large sample sizes [42]. Indeed, inspection of the original overall P-values for each GSV (Table 1: reproduced from or calculated according to the original reports; see Text S1) shows that even a moderate correction to the threshold for significance to account for these multiple tests would have excluded the majority of the reported loci. Thus, our observation of an apparent high rate of false positives is likely to reflect insufficient control of the FDR.
