For each imputed SNP, we obtained a genotype dosage value (a fractional value between 0 and 2 indicating the expected number of minor allele copies) and the most likely discrete genotype, either directly from the output of the imputation software (MaCH, MaCH-Admix, and BEAGLE) or via conversion of the software output (IMPUTE2). For all imputation procedures, we masked 2% of the genotyped SNPs to allow direct comparisons of their true and imputed genotypes. The imputation procedures were repeated 10 times with 10 different sets of randomly masked SNPs, with one exception due to excessive computational runtime (MaCH using the ALL panel, for which only one imputation procedure is presented). Indel variants from 1000 Genomes panels were not included, in order to focus on evaluating the performance of SNP genotype imputation.
