Assessment of genome-wide prediction by using Bayesian regularized neural networks

• Ano de defesa: 2015
• Autor(a) principal: Glória, Leonardo Siqueira
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa
• Palavras-chave em Português: Genômica; Genética - Parâmetros; Redes neurais (Computação); Genética e Melhoramentos dos Animais Domésticos
• Link de acesso: http://www.locus.ufv.br/handle/123456789/6866

Recently there is an increase interest to use nonparametric methods, such as artificial neural networks (ANN). In animal breeding, an especial class of ANN called Bayesian Regularized Neural Network (BRNN) has been preferable since it not demands a priori knowledge of the genetic architecture of the characteristic as assumed by the most used parametric methods (RR-BLUP, Bayes A, B, Cπ, BLASSO). Although BRNN has been shown to be effective for genomic enable prediction. The aim of the present study was to apply the ANN based on Bayesian regularization to genome-enable prediction regarding simulated data sets, to select the most relevant SNP markers by using two proposed methods, to estimate heritabilities for the considered traits, and to compare the results with two traditional methods (RR-BLUP and BLASSO). The simplest Bayesian Regularized Neural Network (BRNN) model gave consistent predictions for both traits, which were similar to the results obtained from the traditional RR-BLUP and BLASSO methods. The SNP importance identification methods based on BRNN proposed here showed correlation values (0.61 and 0.81 for traits 1 and 2, respectively) between true and estimated marker effects higher than the traditional BLASSO (0.55 and 0.71, respectively for traits 1 and 2) method. With respect to h 2 estimates (assuming 0.35 as true value), the simplest BRNN recovered 0.33 for both traits, thus outperforming the RR-BLUP and BLASSO, that, in average, estimated h 2 equal to 0.215.