Genomic prediction of additive and non-additive effects in a pine breeding and simulated population

• Ano de defesa: 2016
• Autor(a) principal: Almeida Filho, Janeo Eustáquio de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa
• Palavras-chave em Português: Genética vegetal; Genômica; Pinus taeda - Seleção; Locos de caracteres quantitativos; Matemática estatística; Genética Quantitativa
• Link de acesso: http://www.locus.ufv.br/handle/123456789/7540

The prediction of individual genetic merit is one of most important challenges in plant and animal breeding. Prediction is difficult because the important traits have a complex nature, where some traits have few genes with major effects, while others are controlled by a large number of genes with small effects. Non-additive effects such as dominance and epistasis can also be important for controlling the genetic variation. In order to achieve higher accuracies in the prediction, it is important to use the model that matches the genetic architecture of trait. The proper partition of the various sources of genetic variation (additive, dominance and epistasis) is desired for several applications, such as exploring the overall and specific combination ability. In Chapter 1, the general remarks of genomic prediction (GP) are reviewed, with the application of this approach with different proposals in distinct genetic architecture traits, together with some statistic models applied in GP. In Chapter 2, the additive and additive-dominance whole-genomic-regression (WGR) models are evaluated with different priors, together with assumptions regarding the presence or not of markers with major effects. Chapter 3 evaluates the inclusion of pedigree information in genomic prediction with additive- and additive-dominance BayesA and also with RKHS model that can theoretically predict confused additive and non- additive effects. These models were applied in tree height (HT), diameter at breast height (DBH) and rust resistance in 923 loblolly pine individuals at 6 years of age from a structured population of 71 full-sib families genotyped with 4722 genetic markers. Six traits were also simulated with distinct genetic architectures (polygenic and oligogenic traits with three dominance levels) for these studies. The simulated population for these traits was derived from a standard pine breeding program. In the oligogenic simulated traits and rust resistance in chapter 2, BayesA and BayesB provided greater accuracies for genotypic prediction; however, the different priors of WGR yielded similar results for HT and simulated polygenic traits. Therefore, the inclusion of dominance effects in WGR increases the accuracy only for simulated traits with high dominance effects and HT. When BayesB was fitted in one generation for predicting the next generation, the dominance inclusion increased the accuracies only for the oligogenic simulated trait with high dominance. Regardless of the model adopted, the accuracy of whole genotypic prediction decreased with the increase of dominance effects in simulated traits. Thus, these results reflect that dominance prediction is complex when compared to additive prediction, and for downstream applications of dominance effects, some genetic properties of the population should be evaluated, such as MAF and the number of half and full-sibs. In chapter 3, the inclusion of pedigree information in genomic model did not yield higher accuracies than models based in only marker information, and both models were substantially more accurate than models basedonly on pedigree. In HT, DBH and in polygenic traits simulated with additive-dominance effects, the RKHS-based models showed slightly higher accuracies than BayesA for whole genotypic prediction, while BayesA-based models were the best option for rust resistance and oligogenic simulated traits. For the prediction of breeding values, the BayesA additive was the best model.