The performance results comparing the 12 imputation strategies (four different programs and three different reference panels from the February 2012 release of 1000 Genomes) are shown in Figure 1 for concordance, Figure 2 for IQS, and Figure 3 for average r2hat. Concordance and IQS served as metrics of imputation accuracy of masked genotyped SNPs, whereas average r2hat served as a metric of imputation quality of all imputed SNPs. Comparisons of the imputation performance results based on HapMap phase III and the August 2010 release of 1000 Genomes (Figures S3, S4, S5) shows that HapMap gave the best imputation performance. However, 1000 Genomes offers substantially denser coverage of the genome, and we found that the February 2012 release provided notably better imputation performance when compared to the earlier August 2010 release.
