Skip to main content

Table 4 Accuracy comparisons between the two-step piecemeal and the classic one-step imputation on the Holstein datasets

From: Whole genome SNP genotype piecemeal imputation

  5-Fold cross validation   Independent testing
BaseProgram Imputation a c c 1 a c c π + #Clusters #TClusters a c c 1 a c c π +
  6 K →50 K 86.98 % 89.81 % 2.87 % 95 189 74.97 % 76.90 % 1.94 %
Beagle 6 K →777 K 82.35 % 85.27 % 2.92 % 1,000 1963 71.29 % 73.25 % 1.96 %
  50 K →777 K 93.09 % 95.16 % 2.07 % 1,000 1956 82.27 % 84.25 % 1.97 %
  6 K →50 K 91.11 % 91.64 % 0.53 % 95 288 81.15 % 81.40 % 0.25 %
FImpute 6 K →777 K 89.22 % 90.14 % 0.92 % 1,000 2942 82.80 % 82.81 % 0.02 %
  50 K →777 K 95.25 % 95.61 % 0.36 % 800 2765 87.72 % 87.83 % 0.11 %
  1. Results are on the Holstein datasets for markers on chromosome 27. Columns 3–7 contain the 5-fold cross validation results on 114 animals, with the selected markers and their associated target marker clusters. Independent testing results on the 8 animals are in columns 8–10, using the selected markers and their associated target marker clusters from the cross validation. In the independent testing, for 1Beagle 6, 37, and 44 target marker clusters are empty; for 2FImpute 7, 58, and 35 target marker clusters are empty. The columns labelled with + show the improvements, in bold, of the piecemeal imputation over the one-step imputation