Our previous work13 suggested potential bias in association testing when using genotypes imputed from the full sets of SNPs from different genotyping arrays. Starting from the intersection of genotyped SNP sets avoids such bias (step 2). We employed the same strategy for merging array and WGS genotypes, but because of the full genome coverage of WGS, the entire set of array SNPs were used. We also investigated the intersection between the WGS data in TOPMed and different targeted arrays (e.g., MetaboChip, Immnochip and OncoArray), and the overlapping rates were all above 95% (Supplementary Table 1), therefore validating the integrating of array and WGS data. The array and WGS data were then independently QC’d using the same QC steps (step 3). This then was followed by phasing, merging, and imputation (steps 4–5). To further reduce potential bias between the array-genotyped and WGS-derived SNPs, a second round of imputation is performed after removing genotyped SNPs with low empirical R2 (ER2 < 0.9, step 6, Supplementary Fig. 1). Finally, following association testing (step 7), filtering based on MAF ( > 0.01), imputation quality
