Having simulated a reference panel, we then simulated a GWAS dataset as the target for imputation. Of the 8400 CEU-like haplotypes from sfs_code that were not included in the reference set, we selected 2000 and randomly paired them to create a GWAS sample of 1000 individuals. We thinned this dataset to the approximate SNP density and MAF distribution observed in real Affymetrix 500k data on a set of British controls [the 1958 Birth Cohort of the Wellcome Trust Case Control Consortium (2007)].
