Predição de valores genéticos utilizando inferência bayesiana e frequentista em dados simulados

Abstract

Simulated data were used to compare EBLUP and Bayesian methods in data with homogeneity of variance, heterogeneity of variance and genetic heterogeneity of genetic and environmental variance. For these structures were strategic disposal of additive genetic and environmental values in accordance with the type of heterogeneity and the desired level of variability: high, medium or low. We used two sizes of population: large and small. For the Bayesian methodology was used three levels of a priori information: no information, just information and informative. For verification of the introduction of different levels of information they were used the mistake percentage in relation to the true value of the variance components the Spearman correlation and the medium square of the mistake among the real genetic values and predicted them. The presence of heterogeneity of variances cause problems for the selection of the best individuals, especially if the heterogeneity is present in the components of genetic variance and environmental and animals are mistakenly selected the more variable environment. The methods presented similar results when compared not informative priors were used, and the populations of large size, in general, showed better prediction of breeding values. Was observed for the Bayesian methodology, the increase in the level of a priori information positively influences the predictions of genetic values, especially for small populations. The Bayesian method is preferred for populations of small size when there is availability of informative priors.