Validating German Holstein single-step evaluations for test-day traits using Interbull’s new GEBVtest software

Abstract

To validate national genomic evaluation systems, particularly those based on a single-step model, Interbull further developed current GEBV test method and extended functionality of the GEBV test python software. As response variable of the linear regression analysis in the GEBV validation test, GEBV as well as deregressed GEBV of validation animals were considered, besides the current deregressed conventional EBV of validation bulls. The aims of this study were to validate the single-step evaluation of four test-day traits for German Holsteins using the newly optimized GEBV test software, and to compare alternative forms of dependent variable and diverse groups of validation animals for genomic validation. Phenotypic, genotypic, and pedigree data were obtained from official April 2021 evaluation for German Holsteins for this study. The single-step evaluations of all the test-day traits were shown to pass the new GEBV test, using dependent variable GEBV or deregressed GEBV for either the validation bulls or cows. For all the tested scenarios, regression slope b₁, genomic model R² and R² increase from a conventional model 2 to genomic model 1 all seemed to meet expectations. Notable variation was observed in the validation results across the subgroups of the validation animals, e.g. the validation bulls born in different years. Dependent variable deregressed GEBV or conventional EBV resulted in clearly lower R² values than GEBV, and the b₁ values deviated slightly more from 1. For the low-reliability validation cows, dependent variables GEBV and deregressed GEBV led to markedly different R² values of the gnomic model 1, though similar R² values were found for the high-reliability validation bulls. The deregressed GEBV seemed to be a more appropriate form of dependent variable for the GEBV test than the dependent variable GEBV, especially for the low-reliability validation cows. The new GEBV test software was proven to work as expected.