Genomic selection in durum wheat: prediction accuracies of FHB resistance in a diversity panel and its application in biparental breeding populations

Citation

Ragupathy, R., Ruan, Y., Berraies, S., Campbell, H., Meyer, B., Cuthbert, R.D., Henriquez, M.A., Kumar, S., Burt, A., Zhang, W., Fobert, P., and Knox, R.E. 2018. Genomic selection in durum wheat: prediction accuracies of FHB resistance in a diversity panel and its application in biparental breeding populations. 9th Canadian Workshop on Fusarium Head Blight/4th Canadian Wheat Symposium, Winnipeg, Canada, November 19-22, 2018.

Résumé

Fusarium head blight (FHB) is a major disease in durum wheat and resistance to FHB involves multiple
loci. Hence, prediction of breeding values using genome-wide markers is a promising strategy for
selection among non-phenotyped individuals and accelerate genetic gain. In the present study, different
statistical models namely, ridge regression-best linear unbiased predictor (rrBLUP), Bayes B and
reproducing kernel Hilbert spaces (RKHS) regression are being evaluated. A durum diversity panel
consisting of 200 individuals, and two doubled haploid populations from the crosses DT707/DT696 and
Strongfield/Blackbird were genotyped using the 90K Infinium iSelect SNP array. The accessions were
evaluated for FHB incidence, severity and index in replicated field nurseries at two locations (Brandon
and Morden) in 2015, 2016 and 2017. Phenotypic data analysis was carried out to estimate least square
(LS) means. From the genotyping assay, a total of 6,899 SNPs were found to be present among the 195
accessions of the durum panel. Using 80% of the 195 lines as the training set, and the remaining 39 lines
as the validation set, predictions were carried out in rrBLUP with 100 independent iterations. The mean
prediction accuracy for FHB index was found to be 0.57 and 0.52 for Brandon and Morden, respectively.
Similarly, the within year genomic prediction accuracy was found to be 0.48, 0.34 and 0.39 for 2015, 2016 and 2017, respectively. Development of two-step genomic selection models using Bayes B and RKHS
algorithms and their implementation in breeding populations as test sets is ongoing, and the results will
be presented.