Legume Genomics and Genetics 2024, Vol.15, No.3, 126-139 http://cropscipublisher.com/index.php/lgg 131 Figure 2 Boxplots of single-sample prediction accuracy of different genomic estimated best linear unbiased prediction (GBLUP) models over 500 samplings of test sets (Adopted from Duhnen et al., 2017) Image caption: Figure 2 displays boxplots of single-sample prediction accuracy of different genomic estimated best linear unbiased prediction (GBLUP) models over 500 samplings of test sets. Each sampling consists of a random split of each subpopulation (early lines or late lines) into a 90% training set and a 10% test set. The models include: All and All_S, which are additive models calibrated on both training sets (early and late), assuming a homogeneous (All) or structured (All_S) population; W_A and W_AI, which are models calibrated on the training set of the subpopulation to be predicted, including additive effects alone (W_A) or together with additive-by-additive epistatic effects (W_AI). As a measure of single-sample prediction accuracy, we used Pearson’s correlation between predicted genetic values and adjusted means. The plots show the prediction accuracy for three traits: yield, seed protein content, and protein yield, in both early and late lines (Adapted from Duhnen et al., 2017)
RkJQdWJsaXNoZXIy MjQ4ODYzNA==