Predição de valores genéticos por abordagens de seleção genômica ampla e de inteligência computacional
Os programas de melhoramento genético existem com dois objetivos principais: identificação de genótipos superiores e a obtenção de combinações melhoradas por meio de cruzamento entre esses indivíduos elite. Os mais diversos ramos da genética, estatística e biometria contribuíram para o estabelecimen...
|Nível de Acesso:||openAccess|
Universidade Federal de Viçosa
|Áreas de Conhecimento:|
|Citação:||SILVA, Gabi Nunes. Predição de valores genéticos por abordagens de seleção genômica ampla e de inteligência computacional. 2018. 108f. Tese (Doutorado em Estatística Aplicada e Biometria) - Universidade Federal de Viçosa, Viçosa. 2018.|
|Resumo inglês:||Genetic breeding programs exist with two main objectives: to identify superior genotypes and to obtain improved combinations through cross-breeding among these elite individuals. The most diverse branches of genetics, statistics and biometry contributed to the establishment of different breeding strategies for selecting superior genotypes. In particular, methodologies based on genomic selection have shown great prominence among the most recent selection studies. Genome Wide Selection, involves biometric studies and gathers genetic of populations, molecular genetics and quantitative genetics. The greatest motivation for such studies is the possibility of using large-scale genotyping and incorporating genomic information into the prediction process, in order to increase selective efficiency, obtain genetic gains and reduce costs. In genetic models, the phenotypic variations of the individuals consist in the genotypic variance including variances due to dominance, environmental variance and also epistasis. However, the GWS models generally neglect the influence of dominance and epistasis, taking into consideration only the additive effects of the characteristics. In addition, the high density of molecular markers can lead to problems of dimensionality and multicollinearity. In this context, the use of dimensionality reduction strategies and methodologies based on computational intelligence that more adequately address the inclusion of effects due to dominance and epistasis in a selection and prediction study are the proposal in this work. The aim of this work is to address three main topics: Chapter ] proposes to evaluate the efficiency of RR-BLUP for predicting genetic values of a simulated population with 12 complex traits that included effects of dominance, epistasis and environmental effects. In Chapter 2 we propose the application of the Stepwise Regression and Sonda methods to reduce dimensionality in order to increase the predictive efficiency of the RR-BLUP method applied in the same population considered in chapter ]. Finally, chapter 3 aims to evaluate the efficiency of computational intelligence methodologies based on Artificial Neural Networks of Multilayer Perceptron and the Radial Basis Function Neural Networks to predict the genetic values of the simulated population discussed in previous chapters. The results indicated that the use of dimensionality reduction methodologies contribute to increase the efficiency of the RR-BLUP method. However, they also evidenced the deficiency of this method to predict genetic values for populations that include effects of dominance and epistasis in the gene control of the characteristics of interest. The methodologies of Multilayer Perceptron and Radial Basis Function Neural Networks proposed presented predictive accuracy, expressed by the mean square error, higher than that presented by the RR-BLUP, demonstrating that the computational intelligence methodologies were more efficient than the Genome Wide Selection for the study of complex characteristics with gene control involving additive, dominant and epistatic effects.|