Fig. 4

Prediction accuracy of different marker sets for five fruit traits in the combined population (n = 353). The eight subsets of SNPs (12,288, 6,144, 3,072, 1,536, 768, 384, 192, and 96) were generated from the confident 31,142 SNPs based on their distributions across 12 chromosomes. Two additional sets (named with GWAS_192 and GWAS_96) were selected based on genome-wide associated study in the combined population. Genomic estimated breeding values (GEBVs) were estimated using the best models: random forest for fruit weight (FW), fruit width (FWt), and pericarp thickness (PT); reproducing kernel Hilbert space (RKHS) for fruit height (FH); and support vector machine for Brix. Prediction accuracy was evaluated based on the Pearson correlation coefficients between GEBVs and observed phenotypes