Predicting Breed Composition in an Angus-Brahman Crossbred Population Using Genomic Data

Breed composition or ancestry information derived from genomic data could be more accurate than data based on pedigree information and could be useful when pedigree information is missing or incomplete. The goal of this study was to examine the feasibility and accuracy of using genomic data to estimate breed composition and identify the minimum number of markers needed to determine breed composition accurately. Some 782 cattle from an Angus-Brahman crossbred population were genotyped with the GGP F250. After applying quality control measures (minor allele frequency, call rate, LD pruning, HWE deviation and genotyping completion), 78,795 SNP and 690 animals remained. Breed membership and level of admixture were estimated using maximum likelihood model based clustering implemented in Admixture v1.3. Principal component analysis was performed to find major axes of variation that efficiently capture population structure. The first principal component (PC1) had strong correlation with pedigree-based breed composition (0.96) and level of admixture inferred by model based clustering (>0.99). Genome wide association was performed with PC1 as a response variable, and SNP were ranked based on the strength of their association with PC1. Based on ranking, 20 subsets were selected with the top 5 up to 100 SNP. For each subset, a 5-fold cross validation was used to assess the accuracy of prediction which was measured as correlation with composition known from pedigree. The results show that few SNP can be used to predict breed composition and accuracies > 0.9 can be reached with as few as 25 SNP.