Modeling unknown parent groups or metafounders in single step genomic BLUP – results of a simulation study

Abstract

The concepts considering for unknown parents are crucial in improving genetic evaluations in animal breeding by accounting for genetic differences within base populations. This study builds on a previous simulation study for the German-Austrian-Czech Fleckvieh population, presenting results that compare metafounders (MF) and unknown parent groups (UPG) for single-step genomic best linear unbiased prediction, and includes detailed analyses for scaling variance components when using MF. The results show that in both settings with complete and incomplete pedigree, evaluations using MF show the best bias and dispersion results, with minimal impact from incomplete pedigree information. In contrast, evaluations without UPG or MF and evaluations where UPG were incorporated via Quaas-Pollak-transformation in the pedigree-based and genomic relationship matrix (UPG_fullQP) exhibit substantial overestimation and overdispersion, emphasizing the importance of accurate relationship modeling in genetic evaluations. This study found that estimating variance components using MF and scaling variance components lead to the same heritability. However, using adapted variance components results in moderate overestimation and slight overdispersion of GEBV. The validation method based on the linear regression method could not detect the significant overestimation and overdispersion in UPG_fullQP. This means that commonly used validation methods tend to underestimate the advantages of MF in populations with numerous unknown pedigrees, highlighting challenges in model optimization for handling unknown parents.