Fig. 2

Predictive ability estimates (PA) for six traits using GBLUP-A models based on 10 repeats of tenfold cross-validations (n = 100) and 904 clone means. a) PAs for traits’ phenotypic values based on GBLUP-A using different numbers of SNPs randomly selected (blue longdashed line), SNPs selected based on lowest p-values in GWAS performed on each training dataset using BLINK method (red twodashed line), and PAs of within family variation (clone mean—family mean) using different number of SNPs based on both randomly selected (purple solid line) and the lowest p-values based on GWAS (green dotted line). The black horizontal dashed line in a) is the square root of narrow-sense clone mean heritability estimated based on the PBLUP-AR model. Matching to GBLUP-A models in a), the standardized Akaike information criterion (AIC) values for each model is shown in b) based on randomly selected SNPs and in c) based on SNPs selected by smallest p-values