Figure 4
From: A joint learning approach for genomic prediction in polyploid grasses
Accuracies of simulation scenarios created using a wheat breeding program and different quantities of QTLs per chromosome (2, 10, 100, and 1000). From 20 years of phenotypic selection (10 years of burn-in), preliminary yield trials (1000 lines) were selected by year, forming a dataset of 10,000 genotypes, which was sampled for different quantities of individuals (100, 250, 500, 1000, 2000, 5000, and 10,000) to perform 10-fold cross-validation using a single-environment main genotypic effect model with a Gaussian kernel (SM-GK) with and without the inclusion of the established feature selection approach (SM-GK/C3R3). The values in the boxplot labels represent the mean accuracy, and in the gray boxes the quantity of markers used to create the predictive model.
