Skip to main content
ARS Home » Plains Area » Clay Center, Nebraska » U.S. Meat Animal Research Center » Genetics and Animal Breeding » Research » Publications at this Location » Publication #294121
Title: Comparison of Bayesian models to estimate direct genomic values in multi-breed commercial beef cattle

Author
item ROLF, MEGAN - Oklahoma State University
item GARRICK, DORIAN - Iowa State University
item FOUNTAIN, TARA - Kansas State University
item RAMEY, HOLLY - University Of Missouri
item WEABER, ROBERT - Kansas State University
item DECKER, JARED - University Of Missouri
item Pollak, Emil
item SCHNABEL, ROBERT - University Of Missouri
item TAYLOR, JEREMY - University Of Missouri

Submitted to: Genetics Selection Evolution
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/4/2015
Publication Date: 4/1/2015
Publication URL: https://handle.nal.usda.gov/10113/60870
Citation: Rolf, M.M., Garrick, D.J., Fountain, T., Ramey, H.R., Weaber, R.L., Decker, J.E., Pollak, E.J., Schnabel, R.D., Taylor, J.F. 2015. Comparison of Bayesian models to estimate direct genomic values in multi-breed commercial beef cattle. Genetics Selection Evolution. 47:23. DOI 10.1186/s12711-015-0106-8.

Interpretive Summary: Despite the preponderance of crossbred animals within the commercial US beef population, few studies have examined the development of multi-breed direct genomic breeding values (DGVs) prediction models using field data. Procedures to generate DGV predictions have been most thoroughly tested in purebred populations due to ease of DNA collection and simple population structure, but these animals do not reflect the levels of admixture and crossbreeding that are present in the commercial beef industry. For these technologies to have the broadest possible impact, we must begin to understand how they can be applied in commercial cattle populations, where pedigree is not known, and admixture is prevalent. The objective of this study was to use a crossbred population of commercial steers and heifers from the National Cattleman’s Beef Association sponsored Carcass Merit Project (CMP) to evaluate the efficacy of DGV prediction models for carcass traits utilizing various proportions of animals in training and validation populations and three different Bayesian prediction models to characterize DGV accuracy.

Technical Abstract: Background Several studies have examined the accuracy of genomic selection both within and across purebred beef or dairy populations. However, the accuracy of direct genomic breeding values (DGVs) has been less well studied in crossbred or admixed cattle populations. We used a population of 3,240 crossbred steers and heifers of mixed breed ancestry produced by five breed associations in the National Cattlemen’s Beef Association (NCBA) Carcass Merit Project (CMP) to predict DGVs for five economically important traits of beef cattle. We compared various sizes of training populations and alternative Bayesian methods for predicting DGVs for carcass traits in admixed populations because DGV accuracy is dependent on population size and genomic architecture. Results Realized accuracies ranged from 0.40 to 0.77, depending on the trait, number of records, and model (BayesA, BayesC0, or BayesCp). Advantages of mixture models (BayesB, BayesC) and models which allow for unequal variances for allele effect distributions (BayesB) were observed for traits such as Warner-Bratzler Shear Force that are influenced by genes of relatively large effect. When genetic architecture of the trait was consistent with the infinitesimal model, differences in prediction accuracy between models were negligible. Randomly assigning 60-70% of animals to training (n ˜ 2,000 records) yielded predicted DGVs with the smallest coefficients of variation compared to larger or smaller fractions assigned to training. Conclusions The training population size (n ˜ 2,000) that optimized accuracy and minimized the coefficient of variation in this study was twice as large as has been recommended as the minimum for purebred cattle evaluations (n = 1,000 for moderate to highly heritable traits). Models fit using BayesA consistently produced high DGV accuracies for all traits; however, the best fitting model depended on the genetic architecture of the trait as defined by the presence of large effect QTL.