New insights on genetic modeling of growth and reproductive traits in tropical crossbred and Nellore cattle.

• Ano de defesa: 2019
• Autor(a) principal: Santana, Talita Estéfani Zunino
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa
• Palavras-chave em Português: Bovinos de corte; Teoria bayesiana de decisão estatística; Genômica - Modelos matemáticos; Aprendizado do computador; Genética e Melhoramento dos Animais Domésticos
• Link de acesso: https://locus.ufv.br//handle/123456789/27154

In genetic evaluations of farm animals, infinitesimal linear models are frequently assumed, which do not consider source of non-additive and non- linear effects, it might reduce the predictive ability, mainly in populations of crossbred animals. In this context, there have been increasing interest in prediction methods that allow access these effects, above all, without assume statistical presuppositions. For predict breeding values in crossbred populations the key point use methods that allow assess non-additive effects (heterosis, complementarity and epistatic losses). However, these effects are highly correlated and frequently assumed as equally relevant. In the sense, a variable selection model (BayesB) was implemented to estimate non-additive effects as well as obtain breeding values for weaning weight in a population with 16,126 beef cattle corresponding to twenty-six crosses compositions. The BayesB proved to be a powerful method to reduce the estimation problems coming from non-additives covariates, and effects frequently assumed as important (maternal non-additive genetic effects and both breed additive effects are not relevant) were statistical reset, opposing the empirical presets assumed in several studies. In addition to benefits statistical promoted by dimensionality reduction, the BayesB model might reduce computational demand and processing time given that enable estimate non-additive effects and predict breeding values in single step, in other words, without additional analysis as it is currently done. It makes the BayesB model very attractive for application in breeding programs of crossbred beef cattle. On the other hand, in the genome-wide selection field, new statistical methods have been proposed in order to minimize the side effects (high-dimensionality and multicollinearity) coming from simultaneous estimation of SNPs. However, the studies applied to genomic classification with machine learning are few. In the sense, the artificial neural network (ANN) methods have been highlighted, however, scenarios with larger genomic data set analyzed by machine learning (ML) algorithms, as ANN, imply in an expensive computational processing. For this reason, searching ML algorithms simplest, was proposed a study of genome-enabled classification by several machine learning frameworks for stayability trait in Nellore cattle. In this study, was performed SNPs selection a set (one, three and five thousand markers), in order to evaluate the impact of structure data set in the classification of daughters. Moreover, was included biological noise in phenotypes in other to challenge to learning algorithms. In this sense, was verify that ML frameworks simplest, as Naïve Bayes, are better to elaborate methods to solve complex issue of classification.