We tested genetic association of all the 1 M SNPs with the cases and controls. A Quantile-Quantile Plot (Q-Q plot) is shown in Figure 2. By comparing the distribution of observed P values against the theoretical model distribution of expected P values, Q-Q plots are used in genome wide association studies to assess the inflation of false positive rates [19]. In randomized data without type I error arising from population stratification or some other artifact, the Q-Q plot should be a 45 degree line. To ensure that our random division of the data did not result in population stratification, we constructed a Q-Q plot based on the true genotypes, which was normal as expected (λ = 1.03) (Fig. 2A). However, the Q-Q plot of imputed SNPs compared to genotyped SNPs is greatly distorted (λ = 1.15), suggesting that combining imputed SNPs with genotyped SNPs without other quality control is problematic (Fig. 2B). Therefore, the observed distortion was due to imputation error and the statistically skewed SNPs (Fig. 2B) are false positives. We then filtered the imputed data by removing all
