Predição de fenótipos de Escherichia coli através de redes biológicas e aprendizado de máquina

• Ano de defesa: 2015
• Autor(a) principal: Reis, Esther Camilo dos [UNESP]
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Sistemas biologicos; Escherichia coli; Aprendizado do computador; Fenotipo; Predição (Logica)
• Link de acesso: http://hdl.handle.net/11449/142957; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/05-07-2016/000865635.pdf

An important question raised after the first complete genome sequencing was: how many genes are essential for the cell life? Single deletion experiments carried out with the bacteria Escherichia coli unveiled that less than 10% of their genes are essential, which means that the inativation of each one leads to the total bacteria inviability. The network theory provides an abstract representation of a biological system, where a set of nodes are the biological components (protein, genes, metabolites, etc) and the set of edges are the interactions (protein-protein physical interactions, metabolic interactions, transcriptional regulational interactions, etc) that link each two biological components. The position of the biological components in a network indicates its importance for the maintenance of the biological system. In general, components located in central positions in a network are those key components for the system integrity. In this work, we decided to survey the position of the 90% genes considered not essential in integrated network of gene interactions (INGI) of the E. coli. Specifically, we investigated the conditionally essential genes, i. e. those genes essential under some type of stress. Moreover, we also investigated the network position of gene pairs that constitute aggravating genetic interaction, i. e. genes pairs that when deleted simultaneously aggravates the organism viability. Using a purely computational approach based on machine learning and topological properties of the INGI, we created preditive decision trees models to define how those conditionally essential genes and the aggravating genetic interaction are distributed in the INGI. A list with the probability of classification for each gene/interaction were obtained. The performance evaluation of our models demonstrates that this methodology can be applied with success in predicting conditionally essential genes. The prediction of genetic interactions also ...