Modelos mistos normais assimétricos em dados de microarrays originados de pedigrees complexos

• Ano de defesa: 2009
• Autor(a) principal: Oliveira, Daniela Carine Ramires de
• Orientador(a): Bueno Filho, Júlio Sílvio de Sousa
• Banca de defesa: Sáfadi, Thelma, Soler, Júlia Maria Pavan, Ferreira, Daniel Furtado, Chaves, Lucas Monteiro
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: UNIVERSIDADE FEDERAL DE LAVRAS
• Programa de Pós-Graduação: DEX - Departamento de Ciências Exatas
• País: BRASIL
• Palavras-chave em Português: Simulação monte carlo via cadeias de markov; Modelo aditivo-dominante; Distribuição normal assimétrica multivariada; Inferência bayesiana; MCMC; Additive-dominance model; Multivariate skew normal distribution; Bayesian inference
• Área do conhecimento CNPq: CNPQ_NÃO_INFORMADO
• Link de acesso: https://repositorio.ufla.br/handle/1/3904

Estimates of heritability for gene expression are scarce and commonly originated from family structures, in which the variability of responses among and within families are provided under a uniform covariance structure for related individuals, ignoring the known relationship among all individuals in the pedigree. Gauss-Markov normal mixed models are the usual choice for such estimates, but in microarrays studies it is common to find asymmetry in residuals of the adjustment of data previously normalized. This, by itself, justifies the use of skew models. In this study it was analyzed a family based pedigree with gene expression measured by microarrays for all individuals. From this pedigree it is possible to estimate additive and dominance variance components and it is reasonable to assume that each of the corresponding individual effects (additive and dominance effects) may have a skew normal distribution. Thus, this work deals with the development and computational implementation of skew normal additive-dominance model for the analysis of microarrays, that allows skewness in all distributions of random effects. Through the MCMC method, it was generated samples from conditional posteriori distributions for all parameters in models with or without skewness parameters for each random effect. It was calculated the Bayes factors for the selection of the best models and HPD intervals for marginal estimates. Results are shown for two of the analyzed probes. For these probes, there was more evidence in favor of models that considered the skew normal distribution for the random effects. The skew normal additive-dominance models considered in this work tended to confound variance components and skewness parameters estimates, possibly due to pedigree limitations. However, these models are the most probable ones and have better residual behavior than their symmetric counterparts.