Skip to main content
ARS Home » Northeast Area » Ithaca, New York » Robert W. Holley Center for Agriculture & Health » Plant, Soil and Nutrition Research » Research » Publications at this Location » Publication #264537
Title: Genomic selection accuracy using multi-family prediction models in a wheat breeding program

Author
item HEFFNER, ELLIOT - Cornell University
item Jannink, Jean-Luc
item SORRELLS, MARK - Cornell University

Submitted to: The Plant Genome
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/24/2011
Publication Date: 3/15/2011
Citation: Heffner, E., Jannink, J., Sorrells, M. 2011. Genomic selection accuracy using multi-family prediction models in a wheat breeding program. The Plant Genome. 4:65-75.

Interpretive Summary: Genomic selection (GS) uses DNA marker data across the genome to predict the performance of experimental lines in breeding programs. In plant breeding, the ability to produce large numbers of lines in single bi-parental families allows GS to be conducted within families. However, this approach requires that the lines from each family be evaluated in the field prior to conducting GS. This field evaluation will prolong the selection cycle and may result in lower gains per year than approaches that estimate marker-effects using performance data from multiple families in previous selection cycles. To test whether estimates from multiple families generated good predictions, we compared them to phenotypic selection (PS) and conventional marker-assisted selection (MAS) on 13 agronomic traits in a population of 374 winter wheat (Triticum aestivum L.) breeding lines. For single traits, average prediction accuracies using GS were 28% greater than with MAS and were 5% less accurate than PS. When combining traits into an index as would be used by a breeder for selection, GS was 14% more accurate than PS. These results provide empirical evidence that multi-family GS could increase genetic gain per unit time and cost in plant breeding.

Technical Abstract: Genomic selection (GS) uses genome-wide molecular marker data to predict the genetic value of selection candidates in breeding programs. In plant breeding, the ability to produce large numbers of progeny per cross allows GS to be conducted within each family. However, this approach requires phenotypes of lines from each cross prior to conducting GS. This will prolong the selection cycle and may result in lower gains per year than approaches that estimate marker-effects with multiple families from previous selection cycles. In this study, phenotypic selection (PS), conventional marker-assisted selection (MAS), and GS prediction accuracy were compared for 13 agronomic traits in a population of 374 winter wheat (Triticum aestivum L.) advanced-cycle breeding lines. A cross-validation approach that trained and validated prediction accuracy across years was used to evaluate effects of model selection, training population size, and marker density in the presence of GxE. The average prediction accuracies using GS were 28% greater than with MAS and were 95% as accurate as PS. For net merit, the average accuracy across six selection indices for GS was 14% greater than for PS. These results provide empirical evidence that multi-family-GS could increase genetic gain per unit time and cost in plant breeding.