Table 2.
Animal‐wise imputation accuracy by scenario
Scenario | Proportion of correctly imputed genotypes (% correct)a | Correlation between true and imputed genotypes (corr)a | ||
---|---|---|---|---|
Average | SD | Average | SD | |
Ref90b | 98.626 | 1.677 | 0.948 | 0.078 |
REf90NoPedb | 98.553 | 1.674 | 0.946 | 0.078 |
Ref50b | 98.390 | 1.819 | 0.939 | 0.088 |
Ref50NoPedb | 98.315 | 1.817 | 0.938 | 0.088 |
Ref10b | 97.432 | 2.359 | 0.916 | 0.095 |
Ref10NoPedb | 97.373 | 2.351 | 0.915 | 0.095 |
RELc | 98.792 | 1.213 | 0.972 | 0.035 |
REL‐Cc | 97.668 | 2.265 | 0.926 | 0.086 |
Statistics were calculated across all 10 replicates for the particular scenarios except for REL, for which there were no replicates.
Dogs were randomly grouped into the reference and the validation sets, and 87.5% of genotypes were masked in the high‐density array to generate the low‐density array; Ref90, 90% of dogs in the reference set; Ref50, 50% of dogs in the reference set; Ref10, 10% of dogs in the reference set; NoPed, indicates that the imputation of the particular variant was run without pedigree information.
Dogs in the reference set (31%) had at least one half‐sibling in the validation set (REL; 69%). The REL‐C controls had the same number of dogs as REL, but dogs were selected at random for the reference and the validation sets. In REL and REL‐C, 87.5% of genotypes were also masked in the high‐density array to generate the low‐density array.