Evaluation of Genomic Selection and Comparison of GS Models for Grain Yield in Soft Red Winter Wheat

Wheat is one of three main cereals feeding the world and the US harvests two billion bushels annually.  Grain yield is a quantitatively inherited trait controlled by multiple genes with small effect genes making phenotypic selection (PS) difficult.  A cycle of PS for yield in winter wheat takes seven years and costs $325 USD per line.  Genomic selection (GS) can shorten the breeding cycle and by combining alleles with small effects to develop breeding values.  A cycle of GS for winter wheat takes one year and costs $75 USD per line.  In GS, genome-wide genotypic and phenotypic data from a population is used to train a model that can be used to to predict breeding values.  The objective of this study was to compare GS models for their ability to predict breeding values and to assess the relative efficiency (RE) of GS compared to PS.  A population of 449 lines was phenotyped for yield (h²=0.53) in five environments and genotyped with 1,820 markers.  The models included ridge regression BLUPS and Bayes A. Cross validation was used to assess the accuracy of GS.  The correlation of phenotypes with RR-BLUPS and Bayes A were 0.38 for both models.  The accuracy of RR-BLUPS and Bayes A was 0.52 using both of the models.  RR-BLUPS and Bayes A appear very similar.  The results show that GS has the potential to be an efficient alternative to PS for improving wheat yield.  Additional data will be presented.