The 1000 Genomes project is being increasingly used in GWAS for SNP genotype imputation, given its highly dense SNP coverage and diverse selection of reference populations. However, there are no well-established strategies to optimize imputation performance, especially in admixed study populations. Our study extends the findings of previous studies evaluating imputation performance using 1000 Genomes reference panels [8], [29], [33], [34], by the following: (1) focusing on the more recent February 2012 release of 1000 Genomes, (2) considering the most diverse ALL reference panel, and (3) making comparisons using recently developed imputation programs for which the ALL panel are specifically advocated. We found that imputation accuracy (based on concordance and IQS results) was comparable across the reference panels. The highest overall imputation quality (based on average r2hat results) was observed for the most closely related YRI+CEU+ASW panel, but this finding was entirely driven by low frequency SNPs, most notably SNPs with MAF≤2%. More specifically, imputation quality based on the most diverse ALL panel was reduced by an abundance of population-specific SNPs that are likely absent in African Americans but present
