Association analysis was conducted using rvTest36 with ten principal components included to account for population substructure. Besides the common filters for minor allele frequency (MAF > 0.01) and imputation quality (R2 > 0.8), we also investigated the imputation quality difference between array-genotyped samples and WGS-genotyped samples by comparing the imputation quality within each sample type, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{array}}^{2}$$\end{document}Rarray2 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{WGS}}^{2}$$\end{document}RWGS2. We verified that the imputation quality between the two types of data were similar. However, some outliers (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{|R}}_{{array}}^{2}-{R}_{{WGS}}^{2}|\ge 0.1$$\end{document}∣Rarray2−RWGS2∣≥0.1) were a major source of false positives, and were removed from the results as a post-association testing filter. Using the same test between COGEND and COPDGene EA sample comparison, inflation of GWAS P-values was apparent when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{|R}}_{{array}}^{2}-{R}_{{WGS}}^{2}|\ge 0.1$$\end{document}∣Rarray2−RWGS2∣≥0.1, but otherwise no inflation was observed (Supplementary Fig. 2a). An imputation quality difference of ≥0.1 only filtered out about 5% of variants (Supplementary Fig. 2b), and the removed variants were scattered throughout the genome (Supplementary Fig. 2c).
