Seleção genômica não paramétrica via distância genética entre subpopulações
Arquivos
Data
2017-02-15
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Viçosa
Resumo
A seleção genômica ampla (Genome Wide Selection – GWS) consiste na análise de um grande número de marcadores SNPs (Single Nucleotide Polymorphisms) amplamente distribuídos no genoma. As principais metodologias propostas e utilizadas na GWS se dividem em metodologias paramétricas, semi-paramétricas ou metodologias de redução de dimensionalidade. Dessa forma, um dos objetivos desse trabalho foi avaliar metodologias não paramétricas, denominadas Delta-p e Regressão Categórica Tripla (TCR), além de compará-las com métodos tradicionalmente aplicados a GWS, tais como G-BLUP (Genomic Best Linear Unbiased Predictor) e BLASSO (Bayesian Least Absolute Shrinkage and Selection Operator). O primeiro capítulo deste trabalho consiste em uma revisão de literatura sobre a GWS apresentando sua definição e importância no melhoramento genético, abordando sobre o desenvolvimento dos métodos propostos e avaliados e também retratando sobre o processo de validação utilizado para a comparação das metodologias. No segundo capítulo, foi proposto e analisado o método Delta-p e um índice de seleção, denominado índice Delta-p/G-BLUP que combina os valores genômicos provenientes do método G-BLUP com os valores genômicos estimados via Delta-p. Sob o contexto Bayesiano, foi incorporado ao LASSO Bayesiano, por meio de uma distribuição a priori altamente informativa, os valores genômicos estimados via G-BLUP, essa abordagem foi denominada método Bayes Híbrido. Para avaliar a eficiência dos métodos estatísticos, no que se refere à estimação dos valores genômicos aditivos e devidos à dominância, foram utilizados dados simulados, sendo estabelecidos oito cenários (dois níveis de herdabilidade × duas arquiteturas genéticas × ausência de dominância e dominância completa) sendo cada cenário simulado dez vezes. Os resultados do segundo capítulo indicaram que o índice Delta-p/G-BLUP e o Bayes Híbrido se mostraram eficientes para predição dos valores genômicos podendo ser usados vantajosamente na GWS. Ademais, no terceiro capítulo, foi avaliada a eficiência do método TCR em comparação com os métodos G-BLUP e BLASSO utilizando quatro cenários (dois níveis de herdabilidade × modelo infinitesimal × ausência de dominância e dominância completa) sendo cada cenário simulado dez vezes. Os resultados indicaram que o método TCR mostrou-se adequado para a estimação dos componentes de variação genômica e da herdabilidade. Em vista disso, uma metodologia baseada em uma modificação do método G-BLUP, denominada TCR/G-BLUP, foi proposta e consiste em estimar a herdabilidade via TCR e fixá-la nas equações de modelos mistos do método G-BLUP. A eficiência dos métodos G- BLUP e TCR/G-BLUP foram comparadas utilizando dados reais, seis caracteristicas avaliadas em mandioca (Manihot esculenta). O experimento foi instalado segundo um delineamento em blocos casualizados com três repetições e 10 plantas por parcela. Os resultados indicaram que o método TCR/G-BLUP foi capaz de aumentar a acurácia e fornecer valores genômicos não viesados se comparados ao método G-BLUP, sendo, portanto recomendado para a aplicação na GWS.
The genomic wide selection (GWS) consists in analyzing of a large number of single nucleotide polymorphisms (SNPs) markers widely distributed in the genome. The main methodologies proposed and used in GWS are divided into parametric methodologies, semi-parametric methodologies or dimensionality reduction methodologies. Thus, one of the objectives of this work was to evaluate non- parametric methodologies, called Delta-p and Triple Categorical Regression (TCR), and to compare them with methods traditionally applied to GWS, such as G-BLUP (Genomic Best Linear Unbiased Predictor) and Bayesian LASSO (Bayesian Least Absolute Shrinkage and Selection Operator). The first chapter of this work consists of a literature review about GWS presenting its definition and importance in genetic improvement, discussing the development of the proposed and evaluated methods and also describing the validation process used to compare the methodologies. In the second chapter, were proposed and analyzed the Delta-p method and a selection index, called the Delta-p / G-BLUP index, combining the genomic values derived from the G-BLUP method with the estimated genomic values via Delta-p. Under the Bayesian context, it was incorporated into the Bayesian LASSO, by means of a highly informative a priori distribution, the genomic values estimated by G-BLUP, this approach was called the Hybrid Bayes method. In order to evaluate the efficiency of the statistical methods, in the estimation of the additive and dominance genomic values, simulated data were used, being established eight scenarios (two levels of heritability × two genetic architectures × absence of dominance and complete dominance) each scenario being simulated ten times. The results of the second chapter indicated that the Delta-p/G-BLUP index and the Hybrid Bayes proved to be efficient for predicting the genomic values and could be advantageously used in GWS. In addition, in the third chapter, the efficiency of the TCR method was evaluated in comparison to the G-BLUP and BLASSO methods using four scenarios (two levels of heritability × infinitesimal model × absence of dominance and complete dominance), each scenario being simulated ten times. The results indicated that the TCR method proved adequate for the estimation of the components of genotype variation and heritability. Therefore, a methodology based on a modification of the G-BLUP method, called TCR/G-BLUP, was proposed and consists of estimating the heritability by means of TCR and fixing it in the mixed model equations of the G-BLUP method. The efficiency of the G-BLUP and TCR/G-BLUP methods were compared using real data, six characteristics evaluated in cassava (Manihot esculenta). The experiment was installed according to a randomized block design with three replicates and 10 plants per plot. The results indicated that the TCR / G-BLUP method was able to increase accuracy and provide non-biased genomic values when compared to the G-BLUP method and is therefore recommended for GWS application.
The genomic wide selection (GWS) consists in analyzing of a large number of single nucleotide polymorphisms (SNPs) markers widely distributed in the genome. The main methodologies proposed and used in GWS are divided into parametric methodologies, semi-parametric methodologies or dimensionality reduction methodologies. Thus, one of the objectives of this work was to evaluate non- parametric methodologies, called Delta-p and Triple Categorical Regression (TCR), and to compare them with methods traditionally applied to GWS, such as G-BLUP (Genomic Best Linear Unbiased Predictor) and Bayesian LASSO (Bayesian Least Absolute Shrinkage and Selection Operator). The first chapter of this work consists of a literature review about GWS presenting its definition and importance in genetic improvement, discussing the development of the proposed and evaluated methods and also describing the validation process used to compare the methodologies. In the second chapter, were proposed and analyzed the Delta-p method and a selection index, called the Delta-p / G-BLUP index, combining the genomic values derived from the G-BLUP method with the estimated genomic values via Delta-p. Under the Bayesian context, it was incorporated into the Bayesian LASSO, by means of a highly informative a priori distribution, the genomic values estimated by G-BLUP, this approach was called the Hybrid Bayes method. In order to evaluate the efficiency of the statistical methods, in the estimation of the additive and dominance genomic values, simulated data were used, being established eight scenarios (two levels of heritability × two genetic architectures × absence of dominance and complete dominance) each scenario being simulated ten times. The results of the second chapter indicated that the Delta-p/G-BLUP index and the Hybrid Bayes proved to be efficient for predicting the genomic values and could be advantageously used in GWS. In addition, in the third chapter, the efficiency of the TCR method was evaluated in comparison to the G-BLUP and BLASSO methods using four scenarios (two levels of heritability × infinitesimal model × absence of dominance and complete dominance), each scenario being simulated ten times. The results indicated that the TCR method proved adequate for the estimation of the components of genotype variation and heritability. Therefore, a methodology based on a modification of the G-BLUP method, called TCR/G-BLUP, was proposed and consists of estimating the heritability by means of TCR and fixing it in the mixed model equations of the G-BLUP method. The efficiency of the G-BLUP and TCR/G-BLUP methods were compared using real data, six characteristics evaluated in cassava (Manihot esculenta). The experiment was installed according to a randomized block design with three replicates and 10 plants per plot. The results indicated that the TCR / G-BLUP method was able to increase accuracy and provide non-biased genomic values when compared to the G-BLUP method and is therefore recommended for GWS application.
Descrição
Palavras-chave
Estatística não-paramétrica, Genômica, Genética de populações
Citação
LIMA, Leísa Pires. Seleção genômica não paramétrica via distância genética entre subpopulações. 2017. 88 f. Dissertação (Mestrado em Estatística Aplicada e Biometria) - Universidade Federal de Viçosa, Viçosa. 2017.