Estimation & Prediction

Using Different Allele Frequencies for Estimating Genomic Relationships and Breeding Values Over Divergent Populations

---

Abstract

Genomic predictions of breeding values within animals involve the construction of a genomic relationship matrix (GRM) to complete a BLUP model. There is interest in understanding how different constructions of the GRM may affect modelling tasks involving multiple and divergent lines of animals, including the bias and dispersion of estimated breeding values and their computed accuracies, and the compatibility of these GRMs with the pedigree-based relationship matrix. The purpose of this work is to analyse the structures of different GRMs and the resulting statistics of GBLUPs applied to a simulated dataset of divergent selection lines. The dataset is a breeding simulation carried out with QMSim. Three divergent selection lines were simulated over several generations starting from a common founder population. Phenotypes and true breeding values were also simulated for analysis of estimating breeding values versus true values and comparing errors against derived accuracies. Genomic relationship matrices were constructed according to Van Raden's method 1 and differentiated by the method for deriving allele frequencies. Genomic BLUPs were fitted with each GRM, and BVs and accuracies were estimated under different settings including the omission of phenotypes from certain generations or subpopulations. Analysis of GRM structure showed that different constructions yielded different distributions of relationships. GRMs based on founder and fixed 0.5 frequencies displayed a monotonically increasing average relationship in later generations, starting from zero at the founders, as in the pedigree-derived matrix. Other GRMs exhibited a saddle structure, placing a reference point not at the founders but at a middle generation. Most GRMs displayed similar statistics (e.g., inbreeding), reflecting expected values, but GRMs from fixed 0.5 frequencies deviated: 0.17-0.27 in inbreeding, 0.51-0.53 in generational, and 0.26-0.40 in parent-progeny means. Without missing phenotypes, EBVs were highly correlated with true breeding values (TBVs) across GRMs (R = 0.951-0.957); notably, EBVs from fixed 0.5 and founder allele frequencies displayed no bias (intercept of EBV regressed on TBV) from TBVs whilst some GRMs generated biases of -1.875 to -2.717. With missing phenotypes, EBVs between subpopulations diverged and EBV-TBV correlations deteriorated to 0.482-0.798 (excluding overall observed frequencies, which produced negative correlations). Accuracies derived from the prediction error variance were generally higher for EBVs from fixed 0.5 frequencies than other matrices. With missing phenotypes, high accuracies (mean 0.848-0.851) were observed for phenotyped individuals and lower accuracies (mean 0.572-0.595) for unphenotyped individuals, compared to 0.902 and 0.756 for fixed 0.5. This work demonstrates that allele-frequency choice when constructing the genomic relationship matrix under a multi-line situation influences matrix structure and its statistics substantially, yielding patterns that do not coincide with the traditional pedigree-derived structure. Furthermore, BLUP EBVs and estimated accuracies varied under different GRM treatments, and curiously produced different accuracies even when estimated BVs were similar.

Keywords: 2026

How to Cite:

Van der Berg, F., Gurman, P. & Henshall, J., (2026) “Using Different Allele Frequencies for Estimating Genomic Relationships and Breeding Values Over Divergent Populations”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286186. doi: https://doi.org/10.31274/wcgalp.23875

Rights: 1