Hanwoo is highly prized for its marbling ability and is the most important cattle breed in Korea. In order to maintain the integrity of the breed and for product certification purposes it is important to develop tools to confirm the origin of the products. Breed composition estimates based on a large number of molecular markers (e.g. HD SNP arrays) are highly accurate but expensive for routine usage. The identification of a reliable panel with a small number of markers will reduce costs and can enable broader adoption of the technology by industry. In this work a heuristic optimization method was used to find the most reliable subset of markers, from the Illumina BovineHD array, to estimate breed proportion in Hanwoo. Accuracies of breed proportion estimates above 90% can be achieved using as little as 200 markers. The best balance between accuracy and number of SNP was obtained with 500 markers achieving 94% accuracy. Rapid and cost effective breed composition prediction in Hanwoo cattle based on a SNP panel with at least 200 markers will help to certify the products with an acceptable accuracy and ensure breed purity within the breeding program. The method described herein is directly applicable to other breeds.

Background: Smallholder dairy farming in much of the developing world is based on the use of crossbred cows that combine local adaptation traits of indigenous breeds with high milk yield potential of exotic dairy breeds. Pedigree recording is rare in such systems which means that it is impossible to make informed breeding decisions. High-density single nucleotide polymorphism (SNP) assays allow accurate estimation of breed composition and parentage assignment but are too expensive for routine application. Our aim was to determine the level of accuracy achieved with low-density SNP assays.

Methods: We constructed subsets of 100 to 1500 SNPs from the 735k-SNP Illumina panel by selecting: (a) on high minor allele frequencies (MAF) in a crossbred population; (b) on large differences in allele frequency between ancestral breeds; (c) at random; or (d) with a differential evolution algorithm. These panels were tested on a dataset of 1933 crossbred dairy cattle from Kenya/Uganda and on crossbred populations from Ethiopia (N = 545) and Tanzania (N = 462). Dairy breed proportions were estimated by using the ADMIXTURE program, a regression approach, and SNP-best linear unbiased prediction, and tested against estimates obtained by ADMIXTURE based on the 735k-SNP panel. Performance for parentage assignment was based on opposing homozygotes which were used to calculate the separation value (sv) between true and false assignments.

Results: Panels of SNPs based on the largest differences in allele frequency between European dairy breeds and a combined Nelore/N'Dama population gave the best predictions of dairy breed proportion ( r² = 0.962 to 0.994 for 100 to 1500 SNPs) with an average absolute bias of 0.026. Panels of SNPs based on the highest MAF in the crossbred population (Kenya/Uganda) gave the most accurate parentage assignments (sv = -1 to 15 for 100 to 1500 SNPs).

Conclusions: Due to the different required properties of SNPs, panels that did well for breed composition did poorly for parentage assignment and vice versa. A combined panel of 400 SNPs was not able to assign parentages correctly, thus we recommend the use of 200 SNPs either for breed proportion prediction or parentage assignment, independently.