We assessed type-I error in a comprehensive analysis involving three smoking-related datasets and their meta-analysis, as shown in Fig. 2b. To fully leverage the large sample size of the COPDGene dataset, we evenly divided the EA samples into two subsets: EA1 and EA2. COPDGene EA1 included all participants diagnosed with COPD (N = 2736) and randomly sampled participants with no COPD (N = 515). The resulting ratio of individuals with COPD in COPDGene EA1 (84%) was close to the ratio in ECLIPSE EA (87%). Three GWAS were conducted to assess type-I error, as follows: (1) array data from COPDGene EA1 (N = 3251) vs. WGS from ECLIPSE EA (N = 1461); (2) array data from COGEND EA (N = 1961) vs. WGS data from COPDGene EA2 (with no COPD, N = 3251); and (3) array data from COGEND AA (N = 712) vs. WGS from COPDGene AA (N = 1710). All association models include ten principal components as covariates to account for population substructure. COPDGene, COGEND, and ECLIPSE are all smoking cohorts and ratios of COPD were consistent across array
