Assessment of alternative models for genetic analysis of worm and tick infestation in nellore cattle

Authors

T. L. Passafaro, University of Wisconsin-Madison
F. B. Lopes, Cobb Vantress Incorporated
Thomas W. Murphy, USDA ARS Roman L. Hruska U.S. Meat Animal Research CenterFollow
B. D. Valente, Genus plc
R. C. Leite, Universidade Federal de Minas Gerais
G. J.M. Rosa, University of Wisconsin-Madison
F. L.B. Toral, Universidade Federal de Minas GeraisFollow

Date of this Version

2-1-2021

Document Type

Article

Citation

Livestock Science 244 (2021) 104276

Comments

U.S. government work

Abstract

Worms and ticks are important parasites in beef cattle, especially in tropical areas, causing significant economic and production losses. Understanding animal-to-animal variation on infestation for these parasites might guide genetic selection and improvement of management practices to attenuate its detrimental effects. Statistical models used to analyze such traits usually assume a Gaussian distribution for the observed data. However, this assumption is quite often inappropriate for counting data. Therefore, the objectives of this study were: 1) Estimate genetic parameters for worms and tick infestations in Nellore cattle, and 2) To compare the overall performance of six data analysis approaches for worm and tick infestation in Nellore cattle, using different specifications of generalized linear mixed models (GLMM) and response variables. Data consisted of presence/absence of parasites as well as counting observations for both worms and ticks in a Nellore herd in Brazil. The binary data were analysed with both Gaussian and Threshold models, whereas the counting data were studied using Gaussian models on the original and logarithmic scales, as well as Poisson and Zero-Inflated Poisson (ZIP) models. All models included the systematic effects of contemporary group and age, as well as the random additive genetic and residual effects. Models were compared using four criteria: Deviance Information Criterion (DIC), Spearman's correlation between predicted breeding values from different models, the agreement on the 5 and 50% top-ranked animals across models, and the Mean Squared Error of Prediction (MSEP) assessed via Monte Carlo Cross-Validation (MCCV). The MCCV was performed using parallel computing through the Center for High Throughput Computing (CHTC) at the University of Wisconsin-Madison. The estimates of heritability ranged from 0.15 to 0.40 for worms and from 0.08 to 0.25 for ticks. According to the DIC, non-Gaussian models displayed the best goodness of fit compared to Gaussian models. DIC's results excluded Gaussian models on the logarithmic scale because fairer comparisons involve phenotypes on the same scale. Spearman's correlation and the percentage of agreement on the 5% and 50% top-ranked animals suggested some re-ranking of animals depending upon the model used. Monte Carlos Cross-Validation showed that all models presented similar MSEP with average values of 0.20 (binary data; worms), 0.18 (binary data; ticks), 15.69 (count data; worms), and 14.19 (count data; ticks). Moreover, performing MCCV in parallel had the benefit to deliver results for all models in about 2 days. Heritability estimates indicate that the selection of high merit animals for worms and tick infestation is possible feasible and can potentially contribute to the genetic progress. Furthermore, genetic selection should be performed concomitantly with traditional parasite control approaches. Overall, non-Gaussian models seem to be better suitable for genetic analysis of worm and tick infestation in beef cattle, because such models have lower DIC values with similar predictive performance compared to Gaussian models.