To assess for the presence of systematic biases in genome-wide analyses, we generated a quantile-quantile(Q-Q) plot of the unweighted p-values (supplemental figure 5). As can be seen in the Q-Q plot, there is no evidence of widespread bias that is generating the associations.To illustrate the distribution of the weights as derived from the GIANT consortium, we provide asimple histogram (figure 1). As can be seen in figure 1, the vast majority of the p-values genome-wide are effectively down-weighted (<1; gray bar, N=522,093) while a fair number of p-values are substantially up-weighted (>1; colored bars, N=195,318). These weights were applied to each of the nominal p-values generated via the linear mixed effect model of longitudinal BMI.The resulting −log10 weighted p-values are displayed using a traditional Manhattan Plot (figure 2). The red horizontal line on figure 2 represents a genome-wide significant threshold (p=5×10−8) while the blue horizontal line represents a threshold of p=5×10−5. Table 4 displays the 39 SNP associations achievingp< 5.0×10−5 ordered by genomic location. We chose a threshold of p< 5.0×10−5 as a reasonable, albeit arbitrary, threshold for association signals
