A nonparametric Bayesian approach for genetic evaluation in animal breeding

Abstract

This article proposes the Bayesian approach to solve problems arising in animal breeding theory. General elements of Bayesian inferences, e.g. prior and posterior distributions, likelihood functions, and the solving of the random effects in the case of the mixed linear model are discussed. Since the random effects are typically assumed to be normally distributed in both the Bayesian and Classical models, a Bayesian procedure is provided which allows these random effects to have a nonparametric Dirichlet process prior distribution. In the case of the Dirichlet process, the Gibbs sampler is introduced to overcome some computational difficulties in solving the genetic parameters of the mixed linear model. To illustrate the application of these techniques, data from the Elsenburg Dormer sheep stud and data from a simulation experiment are utilized.

Genetic and phenotypic parameter estimates of live weight and daily gain traits in Malpura sheep using Bayesian approach

“…The genetic parameter estimates for this breed were obtained earlier by REML approach (Gowane et al, 2010b). However, the Bayesian approach has several practical advantages over the classical (REML) approach (Pretorius and van der Merwe, 2000) like, the estimates from the Bayesian approach for a variance are always positive and an interval estimate such as a highest posterior density region will not include negative values. Therefore, in the current study, major objective was to obtain the estimates using Gibbs sampler animal model.…”

Development of selection strategies for genetic improvement in production traits of Mecheri sheep based on a Bayesian multi trait evaluation

“…However, Bayesian-based Gibbs sampling is a (co)variance component estimation approach that has several practical advantages over the REML method. The Bayesian approach does not specify regularity conditions on the probability model and always returns non-negative estimates for variance and the highest posterior density (HPD) interval [ 13 ]. The posterior probability distributions in Bayesian inference address the issue of uncertainty in unknown parameters and provide more precise estimates [ 6 , 9 ].…”

Development of selection strategies for genetic improvement in production traits of Mecheri sheep based on a Bayesian multi trait evaluation

“…However, Bayesian-based Gibbs sampling is a (co)variance component estimation approach that has several practical advantages over the REML method. The Bayesian approach does not specify regularity conditions on probability model and always returns non-negative estimates for variance and the highest posterior density (HPD) interval [13].…”