Comparing multi-trait Poisson and Gaussian Bayesian models for genetic evaluation of litter traits in pigs

Abstract

a b s t r a c tReproductive traits as number of piglets born (NPB) and weaned (NWP) are directly related to the economic efficiency of swine production systems. Pig breeding programs seek to increase the number of weaned piglets per sow per year, and the NPB is among the factors that directly and indirectly influence the NWP. Thus, multi-trait evaluations are essential to estimate heritabilities and mainly genetic correlations between these traits over different farrowing orders. In general, Gaussian linear mixed models have been used to genetic evaluation of litter traits; however since these traits are characterized as count variables, Poisson models are also indicated. Some studies were carried out using Poisson mixed models in the area of Animal Breeding and Genetics, but they do not point out for a multi-trait scenario, as it should be for litter size at birth and weaning. Toward this orientation, we aimed to apply a multi-trait Poisson mixed model (MPM) for the genetic evaluation of the number of born and weaned piglets under a Bayesian framework. It was aimed also to compare the proposed model with the traditional multi-trait Gaussian model (MGM) by using posterior based goodness-of-fit measures. Two-trait analyses for NPB and NWP were performed separately by each considered farrowing order (first, second and third) using MPM and MGM fitted to data from a commercial Landrace population. Based on DIC (Deviance Information Criterion) and PMP (Posterior Model Probability) values, the MGM outperformed the MPM, but the genetic parameter and breeding values provided by both models were consistent and similar over the three first farrowing orders. Bayesian generalized a multi-trait mixed model approach is feasible for genetic evaluations in the animal breeding context and can be an alternative method for genetic evaluations assuming non-Normal phenotypes.

Bayesian Multi-Trait Analysis Reveals a Useful Tool to Increase Oil Concentration and to Decrease Toxicity in Jatropha curcas L.

“…Implementation of the Bayesian multi-trait models is straightforward, and nowadays it has been widely used due to the possibility of considering a prior knowledge when modeling. Despite its wide application in animal breeding [ 31 , 32 ], Bayesian multi-trait analysis has never been reported in plant breeding.…”

Using Random Regression Models to Genetically Evaluate Functional Longevity Traits in North American Angus Cattle

“…All tested models for each group of culling reason and longevity definition were compared using the Deviance Information Criterion (DIC; [ 26 ]), calculated as follow: in which is the deviance obtained by replacing the parameters by their posterior mean estimates in the likelihood function, and is the effective number of parameters used in the model. In order to facilitate the comparison based on DIC values, the posterior model probability (PMP) was also calculated [ 27 , 28 ]. The PMP is defined as: in which is the posterior probability of the model “ s ” to be the best model; is the exponential of the DIC difference between model “s” and the best model (i.e., the model with the lowest DIC); and is the summation of the exponential of the DIC differences from all tested models, i.e., from the first “s” to the last “S” model.…”

Análise genética da produção <i>in vitro</i> de embriões em bovinos Guzerá

“…Mais recentemente, dados de contagem foram analisados através de Modelos Lineares Mistos Generalizados (Generalised Linear Mixed Models -GLMMs) assumindo distribuição de Poisson (KÖNIG et al, 2007, VAZQUEZ et al, 2009, AYRES et al, 2012ABDOLLAHI-ARPANAHI et al, 2013;VENTURA et al, 2015), binomial negativa GIANOLA, 1999; VARONA; SØRENSEN, 2010) e respectivas distribuições de zero (RODRIGUES-MOTTA et al, 2007;. A teoria para modelos lineares mistos generalizados (GLMMs) foi proposta a partir da percepção que diversas variáveis de interesse seguem distribuição não-Gaussiana.…”

“…As estratégias mais adotadas para modelar resultados da produção de embriões na literatura foram através de transformação da variável , pressuposição de distribuição Gaussiana da variável resposta (TONHATI et al, 1999) e mediante utilização de modelos lineares mistos generalizados . Dados de contagem são comumente modelados assumindo-se distribuições de Poisson (RODRIGUES-MOTTA et al, 2007, VAZQUEZ 2009, AYRES et al, 2012ABDOLLAHI-ARPANAHI et al, 2013;VENTURA et al, 2015)…”

“…Os menores valores observados de MSE para todas as características estudadas quando utilizado o modelo POI indicam que este se adequou aos dados de forma superior aos modelos LIN, LOG e NGB. Alguns autores reportaram comportamento similar PEÑAGARICANO et al, 2010;VAZQUEZ et al, 2012;AYRES et al, 2013), enquanto outros indicaram qualidade de adequação inferior para modelo de Poisson quando comparado ao modelo linear (VAZQUEZ et al, 2009;ABDOLLAHI-ARPANAHI et al, 2013;VENTURA et al, 2015…”