Using Hierarchical Random Regression Model To Optimize The Estimation Of Genetic Parameters Of Milk Production Traits In Dairy Cows

Using test day records to flexibly characterize change in additive genetic effects,random regression test-day model has been successfully applied into genetic evaluation in dairy cattle,which significantly improved precision of estimating genetic parameters and predicting breeding values for lactation traits.To simplify solving random regression model,the Legendre polynomials of different orders were used to fitting the time-depended fixed and random effects.However,choosing high-dimensional random regression models has a heavy computational burden,and fitting the time-depended effects by high-order polynomials maybe produce “Runge phenomenon”.In this study,we stratified the random regression test-day model into two-hierarchical model:the first hierarchy is to model individual lactational curves and test-day effects with a random regression model,and the second hierarchy is to estimate genetic parameters for phenotypic regressions with a multivariate mixed model.For constructing optimal random regression model,residual variance function and what orders of polynomial to fit permanent environmental effects are determined at the first hierarchical model according to BIC criterion,while at the second hierarchical model,what orders of polynomials to fit fixed and additive genetic effects are chosen using analysis of variance for each phenotypic regression and logarithm likelihood ratio of univariate mixed model to its null model for each phenotypic regression.As an application of the hierarchical random regression model,we re-analyzed test-day records for milk production,milk protein and milk fat from 1988 to 1999 in Canadian Holstein dairy cattle.It was concluded that 1)optimal random regression models were constructed,which had the same structures as those chosen with random regression test-day models;2)heterogeneities of residual variances were demonstrated among continuous lactational time intervals;3)Runge phenomenon has not been produced in estimating genetic parameters;and 4)comput ing efficiency was greatly increased to optimize random regression models and estimate genetic parameters...