Imputation-induced bias arises from the union of genotyped SNPs even across similar Illumina arrays (550v1 and 550v3). European American subjects from PanScan were compared to subjects from the CGEMS breast cancer GWAS (Hunter et al. 2007), and African American subjects from two subsets of the iControl database were compared. No spurious SNP associations were found when testing genotyped SNPs available on both arrays or testing SNPs imputed based on the intersection of genotyped SNPs across both arrays (results not shown). Testing SNPs that were imputed based on the union of genotyped SNPs available on either array resulted in spurious SNP associations (Figure S3), albeit the percentages of SNPs showing bias were predictably smaller given the array similarities (0.07 % of imputed SNPs with P < 1 × 10−6 in each ancestry group).
