Abstract
Improving genetic gains in rice breeding programs requires accurate prediction methods for selection indices. Effective use of genomic prediction could significantly accelerate breeding cycles. The Smith index method (SIM), the eigenvalue selection index method (ESIM), and the desired gain index (DG) are linear combinations of trait phenotypic values y (I=b'y), and while the SIM and ESIM predict the net genetics merit (H=w'c), where w is the vector of economic weights and c is the unobserved genotypic values, the DG predicts the mean of genotypic values. To enhance genomic prediction accuracy, mixed linear and Bayesian models incorporate molecular markers to estimate genomic effects, resulting in genomic estimated breeding values. This study evaluated (1) the efficiency of the SIM, ESIM, and DG through their main parameters and (2) the predictive accuracy of 5 genomic prediction models utilizing historical rice (Oryza sativa) data from 2018 to 2021 to predict selection indices for 2022. The correlation between observed and predicted indices assessed the effectiveness of each genomic model. Models incorporating year-specific and environmental covariates significantly improved predictive performance. These findings underscore the importance of environmental covariates and indicate that the SIM is the most effective method for maximizing key index parameters, while the ESIM provides the best predictive accuracy for indices. Consequently, rice breeders are encouraged to use these indices to enhance genetic gains per selection cycle.