Figure 4.

Prediction accuracy of single-trait models for the breeding populations. Prediction accuracy was measured as Pearson’s correlation coefficient (r) between predicted genotypic values and phenotypic values. The prediction accuracy was calculated for all families combined. (A) Five types of validation were compared. Only the mean prediction accuracy of all methods (B) is shown. (B) Twelve methods were tested. RR: ridge kernel regression, GAUSS: Gaussian kernel regression. Validation of type (iii) is shown (A); other validation types are shown in Supplementary Fig. S6. (C) Regression models were based on the results of the single-locus GWAS using parental or combined population. One or three SNPs that showed high −log₁₀(p) values in GWAS were selected for MLR. (i), (iii), and (v) indicate validation types (A). MLR: multiple linear regression. (D) Prediction models that considered only additive or both additive and dominance effects were tested.