Fig 1.
Pairwise genetic dissimilarities between 401 parental lines based on Euclidean distance.
The average clustering method was used to order the lines.
Table 1.
Range and coefficient of variation (CV) of best linear unbiased estimates (BLUE) of genetic effect of genotypes, and variance components of additive genetic effect () and random residual (
), and genomic heritability (
) of mesocotyl length, separately estimated from the parental lines and hybrids populations.
Fig 2.
Prediction accuracies of mesocotyl length using marker-assisted selection with different significance thresholds (P value) for selecting trait-associated SNP markers based on different training set composition scenarios.
The red dash line indicates the prediction accuracy using mid-parental values. The numbers in parentheses indicate the number of effective trait-associated SNPs (TA-SNPs). The asterisks indicate the prediction accuracies based on varying training set composition scenarios were significantly higher (p < 0.05, t-test) than the prediction accuracy using mid-parental values. All prediction accuracies were z-transformed for statistical test.
Fig 3.
Genomic prediction accuracies of mesocotyl length using four prediction models based on different training set composition scenarios.
The red dash line indicates the prediction accuracy using mid-parental values. The asterisks indicate the genomic prediction accuracies based on varying training set composition scenarios were significantly higher (p < 0.05, t-test) than the prediction accuracy using mid-parental values. Different letters above the bars indicated the genomic prediction accuracies of varying scenarios within a specific model were significantly different (p < 0.05, t-test). All prediction accuracies were z-transformed for statistical test.
Fig 4.
Genomic prediction accuracies of mesocotyl length using mid-parental value as a covariate in the genomic prediction models.
Different letters above the bars indicated the genomic prediction accuracies after Fisher’s z-transformation were significantly different (p < 0.05, t-test) between the two scenarios.
Fig 5.
Genomic prediction accuracies of mesocotyl length using four prediction models based on different training set composition scenarios and sizes.
The training set size varied resulted from the alteration of the number of reference hybrids involved in the training set.
Fig 6.
Genomic prediction accuracies of mesocotyl length based on different training set composition scenarios using GBLUP by partitioning SNP markers into trait-associated and -unassociated sets using genome-wide association study (GWAS) based on all lines (A) and all lines and hybrids (B).
The trait-associated SNPs (TA-SNPs) were respectively used as fixed covariates (fixed effect) and independent random kernel (random effect) in the GBLUP model. Undifferentiated use of SNPs was using all SNPs in the GBLUP model. Different letters above the bars indicated the genomic prediction accuracies achieved by varying treatments of SNPs in the model were significantly different (p < 0.05, t-test) after a Fisher’s z-transformation.