Fig. 2

Genotype-to-phenotype (G2P) prediction enhances GWAS power. a Evaluation of the G2P precision using the first prediction scheme of 207 × 27 hybrids as training samples versus the remaining 207 hybrids as testing samples. The red, yellow, and green color scale represents the correlation coefficient (r) between measured and predicted phenotypes from high, moderate, to low r values, respectively. b Evaluation of G2P precision using the second prediction scheme in which 207 × 30 + 1221 Zheng58 F₁ hybrids were used as the training set to predict 1221 Jing724 F₁ hybrids (left panel), and vice versa (right panel). The accuracy of G2P prediction when including parental phenotypes as fixed effects to correct for population stratification was higher than G2P prediction without parental phenotypes. c Comparison of GWAS signals, illustrated as Manhattan plots, in the Zheng58 F₁ population for DTT and mid-parent heterosis of DTT (MPH.DTT; left panel), plant height (PH), and mid-parent heterosis of PH (MPH.PH; right panel) using 207 (“207 measured”) and 1428 samples with measured phenotypes (“1,428 measured”), and 1428 samples with 207 measured plus 1221 predicted phenotypes (“1,428 predicted”). d False discovery rates (FDRs) of GWAS detection of the 20 spike-in QTNs under the circumstances of different prediction accuracy and population size. The GWAS detection for false positives is declared at the significant threshold of p-value ≤ 1e− 05, based on adjusted Bonferroni correction. e Detection power of the 20 spike-in QTNs with major, moderate, and minor effects by GWAS in the population of 1428 samples using the simulated phenotypes at the six levels of prediction accuracy. The GWAS detection of true positives is declared by a significant threshold based on a series of type I errors (α) from 0.05 to 0.95