For quality control of quantitative phenotypes, we excluded (1) related samples (PI_HaT>0.187 estimated by PLINK), (2) samples with age <18 and age >85, and (3) samples with measured values outside three interquartile ranges (IQR) of the upper or lower quartiles. The effect of sex, age, age2, the top 10 genotyping principal components (PCs) and affection status of 47 diseases were removed by linear regression, and the residuals were further normalized by the rank-based inverse normal transformation (see equation (3) below). For quality control of case–control phenotypes, we excluded (1) related samples (PI_HAT>0.187 estimated by PLINK) and (2) samples with age <18 and age >85.
