Modelos de predição genômica multi-ambiental em milho tropical: produtividade de grãos e staygreen

• Autor(a) principal: Crispim Filho, Ailton José
• Data de Publicação: 2023
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFG
• dARK ID: ark:/38995/0013000001tfq
• Texto Completo: http://repositorio.bc.ufg.br/tede/handle/tede/12875
• topic: Seleção genômica; Zea mays L.; Efeito de dominância; Habilidade preditiva; Genomic selection; Dominance effect; Predictive ability; CIENCIAS AGRARIAS

Staygreen and grain yield are agronomic traits of interest to be evaluated in modern maize breeding programs. A modern approach to improving these traits can be genomic selection, whose efficiency depends, among other factors, on the proper choice of the prediction model to be used, the effects that will be accounted for in this model and the resources and time required for the prediction process of the phenotypes. In this work, three parametric models and a non-parametric model were used in the multi-environment genomic prediction of single maize hybrids for staygreen and grain yield, considering additive effects, exclusively, and together with dominance effects. The phenotypic data refer to the evaluation of 152 single maize hybrids, from the crossing of 42 inbred lines, evaluated in 13 environments for grain yield and 8 environments for staygreen. The lines were genotyped with 13,826 SNPs (Single Nucleotide Polymorphism) markers using the GBS (Genotyping by Sequencing) method, and their genotypic combinations were used to generate the genotypes of the hybrids. Adjusted means for each genotype at each location were used to train the genomic prediction models. The predictive ability was measured using Pearson's mean correlation, obtained using the ten-fold system. The models' predictive abilities ranged from 0.23 to 0.83 for grain yield and 0.44 to 0.72 for staygreen. The inclusion of dominance effects in all parametric models increased the predictive abilities for both traits, and for grain yield the average increase was 25%. This confirms that the inclusion of non-additive effects in the prediction model allows better exploration of heterosis and greater precision in genomic selection. The models did not differ between attributes linked to predictive ability. Due to the lower computational demand of GBLUP, it is the most suitable to predict the phenotypic performance of these characters in this data set. Prediction with the additive-dominant GBLUP model indicates the possibility of selecting better combinations of inbred lines than those already performed, which potentially increase grain and staygreen productivity by selecting the best 15 hybrids per prediction for each character separately.