One major, but necessary, obstacle encountered in the current article is the strong multiple test correction. It is tempting to bypass this restriction by conducting targeted analyses of candidate genes, correcting only for the number of tests conducted on those genes. Indeed, targeted sequencing has produced important results in prior work (Bevilacqua et al., 2010), but it remains to be seen whether new sequencing technologies will overcome the known limitations of the candidate gene approach (Hirschhorn, Lohmueller, Byrne, & Hirschhorn, 2002; Sullivan, 2007). Without strong a priori evidence for a candidate gene–phenotype association, and clear genomic function of candidate variants within that gene, we caution that targeted approaches with relaxed statistical or experimental control should be interpreted with circumspection until consistently replicated. If statistical stringency or gene candidacy criteria were relaxed, we would expect a bevy of false-positive association results, and the use of valuable resources to falsify those erroneous findings. The problem is compounded in whole genome sequencing by the large number of protein-coding genes in the human genome and the many different ways to annotate variants within and
