Redes neurais artificiais: novo paradigma para a predição de valores genéticos

• Ano de defesa: 2014
• Autor(a) principal: Silva, Gabi Nunes
• Orientador(a): Cruz, Cosme Damião
• Banca de defesa: Tomaz, Rafael Simões
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado em Estatística Aplicada e Biometria
• Departamento: Estatística Aplicada e Biometria
• País: BR
• Palavras-chave em Português: Engenharia genética; Redes neurais (Computação); Inteligência artificial; Simulação por computador
• Palavras-chave em Inglês: Genetic engineering; Neural networks (Computer); Artificial intelligence; Computer simulation
• Área do conhecimento CNPq: CNPQ::CIENCIAS BIOLOGICAS::GENETICA
• Link de acesso: http://locus.ufv.br/handle/123456789/4079

Until then, how to increase the gain by selection has been pointed out by different strategies recommended in breeding methods, or by using the basic principles of experimentation, or even use biometric models that seek to parameterize both genotypic and environmental influences. Thus, as regards the methods of genetics and statistics used for selection of superior genotypes, highlight the methods derived from the theory of direct selection, indirect selection and index selection, combined selection method and REML-BLUP method among others. Despite several methodologies available for the selection of superior genotypes is still common and necessary practice selection of individuals in relation to features with low reliability of predicting genotype value from a phenotypic value given by the corrected or adjusted by the aggregate average phenotypic information from relatives of correlated traits or practices to reduce the environmental effect. In the end, we consider this phenotypic average adjusted as the most appropriate measure to indicate genetic superiority and predict genetic gain. However, such models or procedures do not include a multitude of other statistical information of great relevance, different phenotypic average which is usually adopted, but which provide important information about the reported genotype and who have been sidelined studies involving genetic improvement and selection criteria. In this context, artificial neural networks are a new paradigm that has been employed, albeit tenuously, in animal and plant breeding programs. This approach, unlike the stochastic modeling used so far, is based on principles of learning and computational intelligence on a wide range of performance information of genotype involving averages, maximum, minimum, variance and all sorts of information can be directly or indirectly measured. Thus, unlike the statistical methods that summarize the information or perform the structural simplification of data, neural networks, like the human brain, capture all available information to generate a criterion for decision making. This work was carried out with the intention of using neural networks to improve the accuracy of the predicted values and genetic gains, through a discussion of its theoretical foundations and use simulated data with the same characteristics in terms of average heritability and coefficient of variation of actual data by providing an alternative method for identification of superior genotypes.