Nordic Holstein single-step test day model using left truncated genomic data

Abstract

In this study, we present the results from the Nordic Holstein test day (TD) evaluation model using left truncated genomic data in the single-step model (ssGTBLUP).  In the truncated genomic data, genotypes of animals born before 2009 were removed from the single-step analyses. It was studied whether the old genotypes can cause bias in the single-step evaluations. The truncated genomic data results were compared to the single-step model with full genomic data using validation where the latest 4 years of phenotypes had been removed. Both analyses used the genomic relationship matrix of VanRaden method 1 and had a 30% residual polygenic proportion (RPG), and an allele frequency of 0.5 for all markers. The results indicate that removing old genotypes reduced the inflation in the young candidate animals considerably, and for validation bulls, the regression (b₁) in predicting the recent GEBV using 4-year-old data improved on average by 11%, and the coefficient of correlation (R²) on average by 5%.  Data truncation had also a positive effect on the differences in the mean Mendelian sampling (MS) terms of young candidate animals.  On the other hand, the truncation of genomic data did not affect the GEBVs of the old, reliably evaluated animals – nor the GEBVs of the old animals whose genotypes were removed, as the within birth year correlation between full genomic GEBVs and genomic data cut GEBVs were nearly 1. Truncation of genomic data removed the over-prediction of recent year classes of bulls and reduced the amount of overdispersion in candidate evaluations to a level acceptable in practice.