Estimation & Prediction

Improving genomic prediction in crossbred dairy cattle using local ancestry

---

Abstract

Crossbreeding has been considered a potential approach to improve production efficiency and sustainability in dairy cattle production system, and its systematic use is of increasing interest on Danish dairy farms. The crossbreds constitute ~10% of the Danish dairy cattle population. The Nordic recording system, with extensive phenotypes and increasing genotyping of crossbred females, supports genomic breeding values for crossbred cows. Incorporating phenotypes from non-genotyped animals and the local ancestry (LA) inference by assigning ancestral origin of chromosome regions can increase accuracy in crossbreds. However, methods and tools for crossbred single-step evaluations that account for LA remain limited. The objective was to evaluate LA-based genomic prediction methods and to prepare a single-step framework integrating purebred and crossbred information. We compared two LA-based approaches: BOM, which uses breed-specific SNP effects weighted by each animal's LA; and BOA, which estimates breed-specific SNP effects jointly from purebred and crossbred data. Three production traits were considered: milk yield (MY), protein yield (PY), and fat yield (FY). The training set included genotypic and phenotypic data up to 2020-01-01, while later animals were used for validation. For within-breed BOM we used data from three Danish dairy breeds, including 48 647 Holstein (HOL), 27 505 Jersey (JER), and 9 992 Red Dairy Cattle (RDC), and for BOA we added 6 517 crossbreds. Prediction was performed on 27 545 HOL, 10 706 JER, 3 732 RDC, and 1 182 crossbreds. For each animal, a predictor of genetic merit (PGM) combined breed-specific SNP effects given LA inference. Predictive ability was the correlation between PGM and corrected phenotype; dispersion bias was the regression slope of corrected phenotype on PGM. Across traits, BOM and BOA had similar accuracies in HOL and JER, while BOA was higher for RDC (up to ~3.8%; e.g. PY: 0.323 vs 0.311). In crossbreds, differences were small but BOA tended to be marginally higher (about 0.9-4.1%; MY: 0.538 vs 0.533; PY: 0.411 vs 0.402; FY: 0.382 vs 0.367). Dispersion bias was generally closer to 1.0 for BOA, indicating better calibration. The results suggest that including crossbred data improved prediction accuracy, but gains may be limited when the number of crossbreds is low relative to the purebreds in the training set. Ongoing work extends these analyses toward an LA-based single-step SNP-BLUP tracing pedigree for five generations (596 777 animals; 134 799 with phenotypes; 178 965 genotyped) based on the animals used in BOM and BOA. Predictions target the same 1 182 crossbreds plus 2 565 additional crossbreds with phenotypes only, to assess gains from integrating non-genotyped animals in a single-step framework. This approach, leveraging pedigree links to non-genotyped relatives, supports accurate selection and mating decisions in commercial crossbred herds.

Keywords: 2026

How to Cite:

Liu, H., Milkevych, V., Su, G., Thomasen, J., Eiríksson, J. & Karaman, E., (2026) “Improving genomic prediction in crossbred dairy cattle using local ancestry”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286822. doi: https://doi.org/10.31274/wcgalp.24133

Rights: 1