The FTND distribution in our sample is nonnormal (Figure 1), but we do not believe this is a major limitation. Linear models assume normally distributed residuals, not traits. Skewed trait distributions may be a concern in GWAS, but the primary risk with such traits is the potential for outliers with extreme values to create spurious results (especially for rare SNPs shared by a small number of individuals with extreme trait values). This is not the case with the FTND symptom count in our sample. The large proportion of the sample with FTND = 0, in fact, limits the potential for this group to produce spurious results for low MAF SNPs. Further, the small genomic inflation factors observed (Figure S1 in Supplement 1) suggest that the trait distribution did not significantly alter the distribution of p values genome-wide. We did, however, correct for the minimal inflation that exists.
