Genetic parameters for milk yield, casein percentage, subclinical mastitis incidence and sexual precocity using Bayesian linear and threshold models

Abstract

Context The economic efficiency of a dairy system is associated with the animal’s productive and reproductive abilities. Therefore, selection criteria should include milk production and quality traits as well as traits related to health and fertility. Since such phenotypes can present non-normal distributions, the use of threshold models is appropriate to study the genetic variation of such traits. Aim To estimate variance components for dairy production and functional traits in a Brazilian Holstein cattle population using linear and threshold models under a Bayesian approach. Methods Data comprised 64 657 test-day records for milk yield (MY, kg/day), casein percentage (CP, % of milk) and subclinical mastitis incidence (SCM), along with 4460 records for sexual precocity (PREC) from 5439 cows. Both SCM and PREC were defined as binary traits. Genetic analyses were based on linear (for MY and CP) and threshold (for SCM and PREC) models using Bayesian estimation. Non-informative and informative priors were considered for variance components, and these models were compared using the deviance information criterion (DIC) and the absolute difference between DIC (Δ). Key results Posterior means of heritability for MY, CP, SCM and PREC were 0.14, 0.39, 0.13 and 0.38 (based on non-informative priors) and 0.13, 0.27, 0.13 and 0.44 (considering informative priors), respectively. The model based on non-informative priors was better (lower DIC) for CP, whereas for PREC, the best model used informative priors. No differences between priors (Δ < 5) were observed for MY and SCM. Conclusions Threshold models were adequate for the analysis of non-normally distributed traits. The use of informative priors can be beneficial if specification is based on results from similar databases and models. Due to their high genetic variation, CP and PREC can be considered as selection criteria in animal breeding programs. In turn, accurate genetic evaluation for MY and SCM will depend on the pedigree and the information from genetically correlated traits. Implications Our study contributes to the understanding of the variation under important dairy production traits in a tropical Holstein population and provides information on the use of Bayesian threshold models as an appropriate method for the evaluation of non-normally distributed phenotypes.