Análise de agrupamento para dados em painel: aplicações em séries temporais de expressão gênica

• Ano de defesa: 2011
• Autor(a) principal: Nascimento, Moysés
• Orientador(a): Sáfadi, Thelma
• Banca de defesa: Bueno Filho, Júlio Sílvio de Sousa, Ferreira, Daniel Furtado, Morais, Augusto Ramalho de, Peternelli, Luiz Alexandre
• 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: Modelo autoregressivo; Microarray time series; Predição da expressão gênica; Séries temporais; Autoregressive model; Gene expression forecasting; Time series
• Área do conhecimento CNPq: CNPQ_NÃO_INFORMADO
• Link de acesso: https://repositorio.ufla.br/handle/1/4401

This work has as objective to propose a methodology for clustering genes with similar expression patterns based on parameter estimates from the autoregressive model of order p, AR (p), for panel data. Were used data from genes expression that are related with the Saccharomyces cerevisiae cell cycle. These data correspond to 114 genes, which each one had 10 fold-change values (expression measure) over time (0, 15, 30, ...., 135 minutes). The parameter estimation of AR(p) panel data model was realized by two different approaches. Under a frequentist approach, we aimed to verify the efficiency of hierarchical (Ward) and optimization (Tocher) clustering methods in the obtaining homogeneous clusters in order to FIT the AR(2) panel data model to gene expression series. The results obtained under this approach indicated that the Ward method was more appropriate to obtain homogeneous cluster in relation to gene expression pattern. Furthermore, the efficiency of the gene expressions forecasting from AR(2) panel data model were 100%. Under a bayesian approach, was considered a joint study of cluster (Ward method) and bayesian analyses of AR (p) model for panel data. The results obtained by this approaches showed that the used methodology provided clustering with similar gene expressions pattern and also effective forecasting for the expression values in future times by predictive distribution theory.