|ZHONG, SHENGQIANG - IOWA STATE UNIV.
|DEKKERS, JACK - IOWA STATE UNIV.
Submitted to: Genetics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/18/2009
Publication Date: 5/2/2009
Citation: Zhong, S., Dekkers, J., Jannink, J. 2009. Association-Based Genomic Selection in Cultivated Barley. Genetics. 182:355-364.
Interpretive Summary: Despite important strides in marker technologies, the use of marker-assisted selection has stagnated for the improvement of quantitative traits. Genomic selection (GS) has been proposed to address deficiencies of current methods. Genomic selection predicts the breeding values of lines in a population by analyzing their phenotypes and high-density marker scores. A key to the success of GS is that it incorporates all marker information in the prediction model, thereby avoiding biased marker effect estimates and capturing more of the variation due to small effect QTL. We evaluated genomic selection using marker data from barley to determine, at marker densities currently available, genomic selection’s viability as a method for barley. We compared the accuracies of four methods as affected by marker density, level of linkage disequilibrium (LD), sample size, and heritability. The GEBV was competitive with the phenotype at predicting true breeding value, without requiring the time and expense of field evaluation. No single method was best in all of our scenarios, however, and we showed what factors were favorable to each method, providing guides for method choice.
Technical Abstract: In genomic selection, the effects of all markers are estimated on a training data set with marker genotypes and trait phenotypes. Genomic estimated breeding values (GEBV) are then calculated for any genotyped individual. We evaluated genomic selection using marker data from barley to determine, at marker densities currently available, genomic selection’s viability as a method for barley. We compared the accuracies of four methods [random regression best linear unbiased prediction (RR-BLUP), Bayes-B, a Bayesian shrinkage regression (BSR) method, and BLUP from a mixed model analysis using a relationship matrix calculated from marker data] as affected by marker density, level of linkage disequilibrium (LD), sample size, and heritability. The GEBV was competitive with the phenotype at predicting true breeding value. The accuracy of GEBV arises from two sources: markers capture QTL effects through LD, and markers contain information on the relatedness of individuals in the training and testing data sets. We found a tradeoff between a method’s ability to capitalize on one versus the other information source, with the BSR method best at the former, Bayes-B competent at both, and the BLUP methods best at the latter. In our study GEBV accuracy derived primarily from the latter source: the BLUP methods were as or more accurate than methods directly estimating marker effects. When markers were in strong LD with large effect QTL, or for predictions on individuals several generations removed from the training dataset, however, rankings of method performance were reversed, suggesting that marker-QTL LD could be important in some cases.