A comparison of nonlinear mixed models and response to selection of tick-infestation on lambs

Abstract

Tick-borne fever (TBF) is stated as one of the main disease challenges in Norwegian sheep farming during the grazing season. TBF is caused by the bacterium Anaplasma phagocytophilum that is transmitted by the tick Ixodes ricinus. A sustainable strategy to control tick-infestation is to breed for genetically robust animals. In order to use selection to genetically improve traits we need reliable estimates of genetic parameters. The standard procedures for estimating variance components assume a Gaussian distribution of the data. However, tick-count data is a discrete variable and, thus, standard procedures using linear models may not be appropriate. Thus, the objectives of this study were twofold: 1) to compare four alternative non-linear models: Poisson, negative binomial, zero-inflated Poisson and zero-inflated negative binomial based on their goodness of fit for quantifying genetic variation, as well as heritability for tick-count and 2) to investigate potential response to selection against tick-count based on truncation selection given the estimated genetic parameters from the best fit model. Our results showed that zero-inflated Poisson was the most parsimonious model for the analysis of tick count data. The resulting estimates of variance components and high heritability (0.32) led us to conclude that genetic determinism is relevant on tick count. A reduction of the breeding values for tick-count by one sire-dam genetic standard deviation on the liability scale will reduce the number of tick counts below an average of 1. An appropriate breeding scheme could control tick-count and, as a consequence, probably reduce TBF in sheep.

Nonlinear mixed‐effects modeling of longitudinal count data: Bayesian inference about median counts based on the marginal zero‐inflated discrete Weibull distribution

“…Table 1 presents the PEs and 95% highest posterior density (HPD) intervals of the bactericidal activity (BA j (0 − 56); see Equation ( 6)) and the regression model parameters. The PEs and 95% HPD intervals of the mean and median log 10 (CFU) counts (see Equation (7)) are shown in Figure 2 by treatment group and day. As expected (due to skewness in the data), the median counts are smaller than the mean counts.…”

“…For example, Klein et al 6 fitted a Bayesian generalized additive model (containing nonlinear smooth functions) based on the ZIP and ZINB distributions to analyze the number of citations of patents granted by the European Patent Office and insurance claims in Belgium during 1997. In a heritability study for tick counts on lambs, Sae‐Lim et al 7 fitted a nonlinear mixed‐effects regression model to the tick counts based on the ZIP and ZINB distributions.…”

Tick Infestation and Piroplasm Infection in Barbarine and Queue Fine de l’Ouest Autochthonous Sheep Breeds in Tunisia, North Africa

“…As an initial Poisson regression of tick count showed overdispersion, a negative binomial regression model was applied [ 51 ]. In this model, the tick count was considered the dependent variable, while both the region and the breed as predictor factors.…”

Suitability of GWAS as a Tool to Discover SNPs Associated with Tick Resistance in Cattle: A Review

“…Additionally, the posterior means of the genetic correlations across the environmental tick infestation surface plot demonstrated a large plateau above 0.80. This indicates that there will be re-ranking of performance for a trait of interest between environments, which necessitates the separation of breeding programs for each environment [105,106]. GxE interaction contributes to the genetic architecture of complex traits and it affects the chances of discovering a true association between phenotype and genotype in GWASs.…”