Genomic prediction models with additive and dominance effects for censored traits

• Ano de defesa: 2017
• Autor(a) principal: Santos, Vinicius Silva dos
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa
• Palavras-chave em Português: Genômica; Análise de sobrevivência (Biometria); Suínos - Genética - Métodos estatísticos; Análise de regressão; Ciências Agrárias
• Link de acesso: https://locus.ufv.br//handle/123456789/27313

Recently, dominance effects have been included in the genomic selection of several species, with the GBLUP-D method being the most used. This method consists in replacing, in the REML / BLUP procedure, the pedigree-based relationship matrices by marker-based relationship matrices. This method can be performed using the GVCBLUP software or through BGLR R-package, which is based on Bayesian regression via the Reproduction Kernel Hilbert Space. The objective of this work was to evaluate the possibility and effectiveness of GBLUP-D implementation via the lmekin function implemented in the coxme package of R through the inclusion of additive and dominance genomic matrices. Thus, through simulated data analyzes, the results obtained by the lmekin function were compared with those obtained by the GVCBLUP software and the BGLR package. Subsequently, the analysis was extended considering phenotypes with censored observations in a F 2 population of pigs, where the time (in days) of the birth to the slaughter of the animal was evaluated through the Cox model and the truncated normal model, in that the censoring was considered or not in the analysis. Finally, the inclusion of the polygenic effect in the additive-dominant models was evaluated in three traits with complete and normally distributed observations of a mice population, and in censored data from a F 2 population of pigs. The results showed that the lmekin function is an efficient alternative for the fit of genomic linear models with additive and dominance effects, since it results were identical to those obtained through GVBLUP software. For the censored data, it was observed a high agreement between the Cox model and the truncated normal model in selecting the best individuals and the highest marker effects. Thus, it was possible to show the possibility of predicting genomic genetic values for censored data, considering the Cox survival model with additive and dominance effects. The inclusion of the polygenic effect in the evaluated models allowed a significative increase in the additive heritabilities of the evaluated traits.