Redes neurais artificiais na predição do valor genético

• Ano de defesa: 2013
• Autor(a) principal: Peixoto, Leonardo de Azevedo
• Orientador(a): Bhering, Leonardo Lopes
• Banca de defesa: Nascimento, Moysés
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado em Genética e Melhoramento
• Departamento: Genética animal; Genética molecular e de microrganismos; Genética quantitativa; Genética vegetal; Me
• País: BR
• Palavras-chave em Português: Redes neurais; Biometria; Análise multivariada
• Palavras-chave em Inglês: Neural networks; Biometrics; Multivariate Analysis
• Área do conhecimento CNPq: CNPQ::CIENCIAS BIOLOGICAS::GENETICA::GENETICA QUANTITATIVA
• Link de acesso: http://locus.ufv.br/handle/123456789/4786

The aim this present work were from the simulated population, make replication or enlargement of populations sets, with the same characteristics off from average, herdability and variation coefficient and organization (covariance or correlation matrix), by means of spectral decomposition, and evaluable efficiency of use artificial neural network in the prediction genetics value in randomized blocks experiments. The design used to simulation of the experiments was randomized blocks with six repetitions. They were simulated 80 conformation, when get established values of average, herdability and experimental variation coefficient. Experiments simulated were used to training (experiment with 5000 genotypes) and validation (100 experiment with 150n and 200 genotypes by each conformation). Thus to measurement the efficiency of the ANN in prediction of genetic value compared its correlation between network value with genetic value and correlation between phenotypic value with genetic value. Average, herdability, variation coefficient and variance and covariance matrix were maintained constant in all experiments simulated. It was possible through spectral decomposition techniques. Neural network were efficient for prediction of genetic value with gain 0,64 until 10,3% in relation phenotypic value regardless of size from populations, herdability or variation coefficient simulated. Thus concluded that preservation variance and covariance matrix of experimental data was efficiently performed by using spectral decomposition of the matrix observed. Therefore datasets simulated can be used to training of the ANNs maintaining original populations characteristics. Thus it was found that ANNs is a technique more efficient by prediction of genetic value in balanced experiments, when compared with phenotype value (average).