Fig. 2
Genomic prediction of agronomic traits in a maize NCII population. A The performance of prediction accuracy by different training datasets and the prediction accuracy between specifically and randomly selected training datasets. In the 5820 F1 hybrids with 194 maternal and 30 paternal lines, the F1s of 1 to 5 diagonal strips were chosen to be the training set, and the remaining F1s acted as the testing set (see Supplementary Fig. 1 for details). The prediction accuracy is evaluated by the Pearson correlation coefficient (r) between the prediction and measured phenotype in the testing set. The red dot indicates the prediction accuracy using three-diagonal strips as the training set (569 individuals). The violin plot indicates the prediction accuracy of 569 randomly selected individuals used as training sets that are repeated 100 times. The significant difference between three-diagonal F1s and random F1s used as training sets is marked by an asterisk for each trait (P<0.01, Student’s t test). B Trait performances achieved by selecting the earliest flowering F1 hybrids. From the testing set of 5251 F1 hybrids, the 100 earliest flowering individuals were selected based on genomic prediction of days to tassel. The early flowering-based selection (red) were compared to the remaining individuals (blue) for days to tassel and ear weight, based on Student’s t test (P<0.01)