Impact of Training Population Composition on Prediction Accuracy: Genomic Selection for Barley Scab Resistance in Upper Midwest

Fusarium head blight epidemics can cause substantial yield losses and quality reductions in malting barley. Although several simulation and validation studies have indicated that breeders could use genomic selection (GS) to improve Fusarium head blight resistance in barley, empirical studies using progeny derived from a training population are needed to design implementation strategies in breeding. To study the effect of training population composition on progeny prediction accuracy, we designed a GS study using six-row spring malting barley germplasm from two closely related breeding programs in the Upper Midwest. For model training, we designed three different panels using 384 breeding lines from the University of Minnesota (M), 384 lines from North Dakota State University (N), and 768 lines from both programs combined (M+N). Our prediction candidates consisted of 300 progeny lines derived from three different cross types from two breeding programs: MxM, NxN and MxN. Each cross type consisted of 100 F₄ derived progeny lines. All lines in the training population were genotyped with 3,072 SNP markers. Prediction candidates were genotyped at the F₃ generation with 384 selected SNP markers and phenotyped for disease resistance. Initial results indicate that (1) genomic selection could be used to predict progeny performance as indicated by previous cross validation studies, (2) prediction accuracy was higher when M was used to predict MxM or N to predict NxN, (3) M+N predicted MxN better than either M or N alone, (4) combining lines from breeding programs to increase training population size did not increase our prediction accuracy.