Optimization of Calibration Sets for Genomic Selection Using Historical Wheat Breeding Data

Implementation of genomic selection (GS) in an applied breeding program may expedite genetic gain. Genomic Estimated Breeding Values (GEBV) were calculated using the statistical model, RR-BLUP (Ridge Regression Best Linear Unbiased Predictor), with genome-wide marker information. It was possible to gain insight into the utility of GS and to optimize calibration sets for the target population of environments. Historical data extending to 2006 was used in the CSU wheat breeding program to optimize calibration sets according to predictive accuracy, as indicated by a correlation, r, between GEBV and observed values for traits - yield, test weight, plant height, and heading date. Wheat genotypes included released cultivars and advanced breeding lines, tested in multi-year and multi-locational replicated yield trials, assayed using genotyping by sequencing (GBS) to obtain about 45k markers. The software, R-language, was used for data analysis with the R-package, rrBLUP. Calibration sets comprised lines grown from one to seven years for predicting breeding values of a validation set, comprising lines grown during one year. Optimal calibration sets differed for each of the traits. For grain yield, calibration sets comprising lines grown during one year were at least as accurate as sets of lines grown for more years – accuracies averaged 0.70. Results of optimization and accompanying co-variables – heritability, relatedness and CDmean (mean of generalized coefficient of determination) - will be presented.