Third, as might be expected given our modest statistical power, none of our results reached levels of statistical significance that would provide high confidence that individually they represented true positive findings. Given this low power and the large number of statistical tests performed, it is plausible that a substantial proportion or even all of our results reflect false positives. This problem is not unique to our study and has been seen with other published findings with complex behavior traits including the prior published GWAS studies of AD (Edenberg et al. 2010; Johnson et al. 2006; Treutlein et al. 2009). Our own judgment is that a number of our results are suggestive and in aggregate unlikely to reflect only random effects. Here, we extended our analyses through gene-wise permutations which led to a number of significant results and importantly, these were for genes which represent plausible candidates based on earlier work. In all, we conducted several levels of analyses to examine enrichment of signals and produced statistically significant results at the gene-level, and a few at experiment-wide level, for signals affecting
