Abstract:

Objective: Genomic selection for optimizing dynamic growth traits during the seedling stage of poplar is crucial for improving prediction accuracy and facilitating early selection of superior offspring. Method: A total of 400 F₁ hybrid progeny derived from a cross between Nanlin 895 poplar (female parent) and Jingxing Yihao poplar (male parent) were used as materials. Ground diameter and plant height were measured monthly from April to September, and genotypic data were obtained through whole-genome resequencing. Six genomic selection models, such as GBLUP, BayesA, BayesC, support vector regression, gradient boosting, and random forest, were employed to evaluate the impact of monthly phenotypic data on the prediction accuracy of genomic selection. Additionally, final ground diameter and plant height were measured at the end of the growing season in December to validate the predictive accuracy of each genomic selection model for final phenotypic values. Result: The mean values of ground diameter and plant height in the hybrid population increased gradually over the months, reaching their maximum in September. The coefficients of variation ranged from 0.23 to 0.34 for ground diameter and from 0.18 to 0.54 for plant height, indicating substantial genetic variation and significant selection potential. Narrow-sense heritability estimates ranged from 0.42 to 0.47 for ground diameter and from 0.39 to 0.62 for plant height. Among the genomic selection models, GBLUP, BayesA, BayesC, and support vector regression consistently had higher prediction accuracy for ground diameter and plant height across all months compared to gradient boosting and random forest. The highest prediction accuracies for ground diameter and plant height were observed in June and September, respectively. The final growth data collected in December was used to evaluate the prediction accuracy of different genomic selection models, and the results showed that models constructed with phenotypic data from June, July, August, and September had significantly higher prediction accuracies for ground diameter and plant height compared to those built with data from April and May. Among them, the BayesA model based on September phenotypes exhibited the highest prediction accuracy for the both traits and was therefore selected to predict and screen the breeding values of the 400 hybrid progenies. Based on the December phenotypic observations and the predicted breeding values through genome selection, four hybrid progenies were consistently selected by both methods for their superior genotype. Conclusion: Genomic selection can effectively identify superior progeny for dynamic growth traits during the seedling stage of poplar, providing an efficient method for the early selection of superior progeny in poplar breeding.

Key words: genomic selection, dynamic growth traits, poplar, prediction accuracy, five-fold cross-validation