Estatística Aplicada e Biometria

URI permanente para esta coleçãohttps://locus.ufv.br/handle/123456789/195

Navegar

Resultados da Pesquisa

Agora exibindo 1 - 1 de 1
  • Imagem de Miniatura
    Item
    Ridge, lasso and bayesian additive-dominance genomic models and new estimators for the experimental accuracy of genome selection
    (Universidade Federal de Viçosa, 2015-10-26) Azevedo, Camila Ferreira; Resende, Marcos Deon Vilela de; http://lattes.cnpq.br/8861113007032888
    The main contribution of molecular genetics is the direct use of DNA information to identify genetically superior individuals. Under this approach, genome-wide selection (GWS) can be used with this purpose. GWS consists in analyzing of a large number of SNP markers widely distributed in the genome. This simulation work presents a complete approach for genomic selection by using adequate genetic models including dominance effects, which are essential for selecting crosses and clones as well as for improving the estimation of additive effects for parent selection. To date, the approaches via Ridge, Lasso and Bayesian additive-dominance models have not been evaluated and compared in the literature.The performance of 10 additive-dominance prediction models (including current ones and proposed modifications) were evaluated. A new modified Bayesian/Lasso method (called BayesA*B* or t-BLASSO) performed best in the prediction of genomic breeding value of individuals, in all the four scenarios (two heritabilities × two genetic architectures). The BayesA*B*-type methods showed better ability for recovering the dominance variance/additive variance ratio. Also, the role of the three quantitative genetics information sources (called linkage disequilibrium, co- segregation and pedigree relationships) in genomic selection were elucidated by decomposing the heritability and accuracy in the three components and showing their relations with the structure of populations and the genetic improvement in the short and long run. Moreover, this simulation work also, we developed the new estimators for the prediction accuracy of genomic selection. The work proposes and evaluates the performance and efficiency of these new estimators called regularized estimator (RE) and hybrid estimator (HE). The regularized estimator takes in consideration both the genomic and trait heritabilities, in addition to the predictive ability. The hybrid estimator (HE), combines both experimental and expected accuracies. The comparisons of the RE and HE with the traditional (TE) were done under four validation procedures. In general, the new estimator presented accuracies closer to the parametric ones, mainly when selecting markers. It was also less biased and more precise, with smaller standard deviations than the traditional estimator. The TE can be used only with independent validation, where it tends to perform better than RE, although overestimating the accuracy. The hybrid estimator (HE) proved to be very effective in the absence of validation. The independent validation showed to be superior over the Jacknife procedures, chasing better the parametric accuracy with or without marker selection. The following inferences can be made according to the accuracy estimator and kind of validation: (i) most probable accuracy: HE without validation; (ii) highest possible accuracy: TE with independent validation; (iii) lowest possible accuracy: RE with independent validation.