Inferência de redes de regulação gênica utilizando métodos de busca e otimização

• Ano de defesa: 2016
• Autor(a) principal: Hattori, Leandro Takeshi
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Elétrica e Informática Industrial; UTFPR
• Palavras-chave em Português: Bioinformática; Algorítmos genéticos; Engenharia elétrica; Genética; Bioinformatics; Genetic algorithms; Electric engineering; Genetics; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/1837

For better understanding the mechanics of cellular control, many different approaches have been developed for inferring Gene Regulatory Networks (GRNs), using temporal gene expression data. However, the large amount of genes observed in contrast with the small amount of gene expression samples makes the inference of GRNs one of the most important problems in Bioinformatics. In this dissertation, the inference of GRNs is a problem decomposed in n feature selection sub-problems. For each sub-problem, the predictor genes for each target gene are obtained. Basically, the feature selection method is composed by a search algorithm and a criterion function. In this work we used bioinspired methods (DE, BAT and ABC) and sequential search methods (SFS and SFFS), and the criterion function we used the Mean Conditional Entropy (MCE). Also, we proposed some pos-processing methods for the optimization of the GRN inferred by the bioinspired methods, by using the Quine-McCluskey (QM) algorithm as well as a consensus network generated from the networks inferred by the bioinspired methods and later optimized by the QM. For the inference experiments, we explored Artificial Genic Networks (AGN) based on Probabilistic Boolean Networks (PBNs), changing the features of the topology, the average number of connections, the number of genes, and the amount of gene expression data. Results showed that the DE algorithm obtained better results, regarding accuracy, when compared with the sequential search methods. When compared with the other bioinspired methods, DE also achieved better results than BAT and ABC. For the optimization of the inferred networks by the bioinspired methods, the QM algorithm presented a good performance, removing predictor genes that were not contained in the real network, leading to an improvement of the accuracy of the inferred network, and its similarity with the real one. The consensus network presented accuracy results and similarity even better than those obtained by the bioinspired methods alone. Overall results suggest that the application of the consensus method based on the bioinspired methods together with the QM pos-processing is promising for the GRNs inference problem.