Optimizing the Training Population to Improve Genomic Selection Prediction Accuracy Across Three Cycles of Selection in a Barley Breeding Population.

Genomic selection (GS) is fast becoming a ubiquitous concept in plant breeding. Empirical validation studies are beginning to testify to the effectiveness of the selection theory proposed over a decade ago. To date, the data and anecdotal experience of users indicates that the accuracy of GS is greatly influenced by the composition and size of the training dataset, or population (TP). The University of Minnesota spring barley breeding program implemented GS in 2009, and has since developed a balanced experimental dataset including 50 progeny selected via GS and 50 progeny selected at random from each of the first three cycles of selection. The TP used for the first three cycles consists of ~1,100 breeding lines from three spring six-rowed barley breeding programs, tested in unbalanced field trials over a four year period. The experimental set, which will serve as a validation population (VP), has been grown in five yield trials and six disease nurseries. Using this dataset we will i) evaluate multiple algorithms for their ability to refine the size and composition of the initial TP, ii) investigate the effect of introducing related individuals into the TP by progressively incorporating progeny from subsequent breeding cycles into the prediction of later breeding cycles, and iii) a combination of i) and ii).