We used the MaCH program to identify stretches of haplotypes shared between the study samples and the HapMap or 1000G SNPs reference panel and to impute their SNPs into the study sample. In both cases, SNP imputation achieved high estimated imputation accuracy, with 2,492,059 and 7,432,030 SNPs imputed with R2 > 0.3, respectively, in the MRCA data set, and 2,429,403 and 7,378,292 SNPs, respectively, in the MRCE data set with R2 > 0.3. (The R2 measure estimates the correlation between imputed and true genotypes, based on the residual uncertainty in estimated genotypes.) We further assessed the accuracy of genotype imputation by comparing imputed and experimentally derived genotypes on a genomic scale. In the first sample (MRCA), individuals were genotyped on two platforms (Illumina 300K or ILMN300K and Illumina 100K or ILMN100K). The ILMN300K genotypes were used to drive imputation. We then used genotypes for markers in the ILMN100K panel but not present in the ILMN300K panel to assess the accuracy of imputed genotypes. Accuracy was measured by correlation between true allele counts and imputed allele counts.
