Comparison between original and modified MaCH.

Comparison between original MaCH and a modified version with our remedy, showing both the total number of switch errors and the number of incorrectly imputed alleles. The comparison is based on the 30 first phased Hapmap3 release 2 CEU trio parents [13]. Four versions are used: 1. the original dataset (only parents), 2. including their children, as well as 3. simulating siblings to parents, 4. simulating twins to parents. When children are excluded and no virtual siblings are present, no known relationships exist between the individuals in the dataset. Imputation performance was verified by reconstructing the half of the marker set () that was left out, using minimac [11], employing the remainder of the phased CEU trio data (57 individuals) as reference panel. All MaCH runs were executed for iterations, with rounds for minimac. Metrics are reported for only the original individuals, in order to aid comparisons.
*In this case, the minimac run starting from the recombination frequencies determined by the original MaCH failed to converge at all, with errors for all markers. The results for original MaCH in this table row are based on the pre-phased haplotypes from original MaCH, but starting out with the recombination frequencies from the modified version, in order to allow the minimac imputation to complete at all.