Estimation & Prediction

Optimizing genetic group identification for genomic evaluation in an admixed population

---

Abstract

Abstract Text: Genomic selection is widely practiced in livestock production systems, as it improves the prediction accuracy compared to traditional pedigree-based genetic evaluation. With the widespread use of crossbreeding programs, it is important to account for breed effects to ensure unbiased prediction of breeding values. This study aimed to simulate an admixed population to assess the use of clustering, principal components, and breed proportion in genomic prediction. A simulation was performed using QMSim to generate a historical sheep population with a genome consisting of 26 autosomal chromosomes and 49,400 SNPs. A three-breed crossbreeding system (populations A, B, and C) was simulated from the historical population using a customised R script. Selection was applied to a different trait in each population, all with a heritability of 0.3. Haplotypes of the recent populations were coded independently and used to estimate true breed proportion (TBP) for each animal. One thousand sheep were simulated per generation with an equal sex ratio. All 500 dams were selected along with 20, 40, and 60 sires based on true breeding value in populations A, B, and C, respectively. Populations A and B were simulated for 5 generations, followed by a cross in generation 6 (AB cross), with further intercrossing until generation 9. Population C was simulated through generation 9, then crossed with the AB population in generation 10 (ABC cross), followed by intercrossing until generation 15. The genotype information of the animals from generations 9 to 15 (N = 8,000) was used for genomic best linear unbiased prediction (GBLUP) analyses, where animals from generations 9 to 14 (N = 7,000) served as the reference population and animals from generation 15 (N = 1,000) as the target population used for validation. A GBLUP analysis without group effects was used as a baseline to compare models fitting group effects as fixed effects. Categorical effects included a pedigree-based groups (PG) from simulation and hierarchical clustering on principal components (HCPC). The HCPC method uses principal components of the genomic relationship matrix (GRM). The models analyzed included first and second principal components, HCPC, PG, TBP, and breed proportions derived from Admixture. Prediction accuracy was estimated as the correlation of genomic estimated breeding value and adjusted phenotype for fixed effects divided by the square root of heritability in a forward validation. The results revealed that the PG model, when fitting groups, achieved the same prediction accuracy (0.47) compared to the base GBLUP model without genetic groups. Other genomic models didn't improve the prediction accuracy. These findings suggest that the GRM sufficiently captures the population structure of the admixed population with available genotypes on all animals.

Keywords: 2026

How to Cite:

Sahoo, S., Ferdosi, M., van der Werf, J. & de las Heras-Saldana, S., (2026) “Optimizing genetic group identification for genomic evaluation in an admixed population”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286175. doi: https://doi.org/10.31274/wcgalp.23869

Rights: 1