O uso de simulação de Monte Carlo via cadeias de Markov no melhoramento genético

• Autor(a) principal: Nascimento, Moysés
• Data de Publicação: 2009
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/4022
• topic: Simulação estocástica; MCMC; Estatística genômica; Inferência bayesiana; Stochastic simulation; MCMC; Genomic statistics; Bayesian inference; CNPQ::CIENCIAS AGRARIAS

The objective of this work was to provide a theoretical and applied reference on the main Monte Carlo simulation methods via Markov chains (MCMC), seeking to focus on applications in genetic breeding. Thus, the algorithms of Metropolis-Hastings, simulated annealing and the Gibbs sampler were presented. The theoretical aspects of the methods were approached through a detailed discussion about their foundations based on the Markov chain theory. Besides the theoretical discussion, concrete applications were developed. The Metropolis-Hastings algorithm was used to achieve estimates from the frequencies of recombination between pairs of markers of a population F2, of co-dominant nature, with 200 individuals. The simulated annealing was applied to establish a better linking order in the construction of genetic maps of three simulated populations F2, with markers of co-dominant nature, containing 50, 100 and 200 individuals, respectively. For each population, it was established a genome with four linking groups, each with 100 cM of size. The linking groups present 51, 21, 11 and 6 markers, with a distance of 2, 5, 10 and 20 cM between the adjacent marks, respectively, providing different degrees of saturation. The Gibbs sampler, on the other hand, was used for the achievement of the estimates of the adaptability and stability parameters of the model proposed by Finlay and Wilkinson (1963), through the Bayesian inference. The data of the productivity averages of five genotypes evaluated in nine environments were used, come from essays in randomized blocks with four replications. In all the applications, the algorithms were computationally viable and achieved satisfactory results.