Fig. 5

Heatmap showing the average increase (orange) or decrease (blue) of Spearman correlation between test set genotypic values and GEGVs generated by GBM model for multiple training set optimization methods relative to using the entire candidate set to train the model. The average Spearman correlation change is calculated for each trait (displayed on the right-hand side of the vertical axis), optimization method (displayed on the bottom of the horizontal axis), and test set (displayed on the top of the horizontal axis) across repetitions and years included in the candidate set. The Spearman correlation was calculated in several subsets of the test set, created by selecting the highest/lowest genotypic values for the trait of interest (left axis). It is noteworthy that the training set size used for all methods was optimized previously by Min_GRM, except for Tails_GEGVs_sd1, which concurrently optimized the training set size and composition