Estimation & Prediction

On the nature of EBV accuracies and reliabilities under multi breed analyses

---

Abstract

It is commonplace for accuracies or reliabilities to be presented to animal breeders as a metric of confidence in prediction and of expected changes in an EBV with additional information. The Australian sheep national genetic evaluations are multi breed and the beef evaluations are currently being developed to allow for multiple breeds to be analysed simultaneously. The question of accuracies under multi breed is more complex, with the problem often avoided in routine evaluations. Approximation algorithms designed to calculate accuracies avoid the extra complexity of unknown parent groups / genetic groups. Accuracies calculated from a metafounder model have received recent attention, due to the lower accuracies predicted compared to a model without groups considered. This study examines different models for estimating prediction error variance based accuracies of EBVs where multiple subpopulations are analysed together. Pedigree, genotypes and phenotypes were simulated with QMSim with the heritability set at 0.3 for three divergent populations: Pop1 minimised inbreeding, Pop2 maximised inbreeding and Pop3 employed random mating. The three populations were simulated for 15, 25 and 30 generations, with the maximum Fst value between populations being 0.20. The first 10 generations of information, pedigree, genotypes and phenotypes were discarded to disconnect the subpopulations. Prediction error variances were then calculated for models without genetic groups, with both implicit and explicit genetic groups calculated as qg+u, showing equivalence of reliabilities between these models, and metafounders. Two metafounder gamma matrices were examined: one calculated based on the covariance between the allele frequencies for generation 10 (MF), and one based on the relative difference in gamma between generation 10 and zero (MF relative). Mixed model equations were inverted using python and cholmod. The mean reliabilities for each population and model are presented in the table. Of note is that the accuracies from the model with genetic groups including the group component is higher than the other models and the metafounder model is much lower, while all other accuracies are similar. Moreover, while all the other models had accuracies ranging between 0 and 1, the accuracies of the full EBV from the genetic groups model had a minimum of 0.33. This observation has relevance if accuracies are used for filtering EBVs for reportability purposes, indicating changes in reportability thresholds may be needed. Accuracies may be more complex to calculate than EBVs, but calculating accuracies that capture subpopulation influences on EBVs is vital for breeders to make informed decisions for selection across subpopulations.

Keywords: 2026

How to Cite:

Gurman, P., Van der Berg, F., Alexandri, P. & Henshall, J., (2026) “On the nature of EBV accuracies and reliabilities under multi breed analyses”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286287. doi: https://doi.org/10.31274/wcgalp.23907

Rights: 1