Influence of multi-trait modeling, dominance, and population structure in genomic prediction of maize hybrids

• Autor(a) principal: Lyra, Danilo Hottis
• Data de Publicação: 2017
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: http://www.teses.usp.br/teses/disponiveis/11/11137/tde-22032018-131222/
• topic: Copy number variation; Efeitos não-aditivos; Gaussian kernel; Kernel Gaussiano; Milho tropical; Non-additive effects; Tropical maize; Variação no número de cópia

Genomic prediction of single-crosses is a promising tool in maize breeding, increasing genetics gains and reducing selection time. A strategy that can increase accuracy is applying multiple-trait genomic prediction using selection indices, which take into account the performance under optimal and stress conditions. Moreover, factors such as dominance, structural variants, and population structure can influence the accuracy of estimates of genomic breeding values (GEBV). Therefore, the objectives were to apply genomic prediction (i) including multi-trait models, (ii) incorporating dominance deviation and copy number variation effects, and (iii) controlling population structure in maize hybrids. Hence, we used two maize datasets (HELIX and USP), consisting of 452 and 906 maize single-crosses. The traits evaluated were grain yield, plant and ear height, stay green, and four selection indices. From multi-trait GBLUP and GK, using the combination of selection indices in MTGP is a viable alternative, increasing the selective accuracy. Furthermore, our results suggest that the best approach is predicting hybrids including dominance deviation, mainly for complex traits. We also observed including copy number variation effects seems to be suitable, due to the increase of prediction accuracies and reduction of model bias. On the other hand, adding four different sets of population structure as fixed covariates to GBLUP did not improve the prediction accuracy for grain yield and plant height. However, using nonmetric multidimensional scaling dimensions and fineSTRUCTURE group clustering increased reliability of the GEBV for GY and PH, respectively.