Predicting breeding values for milk yield of Guzera (Bos indicus) cows using random regression models

Abstract

Given the importance of Guzera breeding programs for milk production in the tropics, the objective of this study was to compare alternative random regression models for estimation of genetic parameters and prediction of breeding values. Test-day milk yields records (TDR) were collected monthly, in a maximum of 10 measurements. The database included 20,524 records of first lactation from 2816 Guzera cows. TDR data were analyzed by random regression models (RRM) considering additive genetic, permanent environmental and residual effects as random and the effects of contemporary group (CG), calving age as a covariate (linear and quadratic effects) and mean lactation curve as fixed. The genetic additive and permanent environmental effects were modeled by RRM using Wilmink, All and Schaeffer and cubic B-spline functions as well as Legendre polynomials. Residual variances were considered as heterogeneous classes, grouped differently according to the model used. Multi-trait analysis using finite-dimensional models (FDM) for testday milk records (TDR) and a single-trait model for 305-days milk yields (default) using the restricted maximum likelihood method were also carried out as further comparisons. Through the statistical criteria adopted, the best RRM was the one that used the cubic B-spline function with five random regression coefficients for the genetic additive and permanent environmental effects. However, the models using the Ali and Schaeffer function or Legendre polynomials with second and fifth order for, respectively, the additive genetic and permanent environmental effects can be adopted, as little variation was observed in the genetic parameter estimates compared to those estimated by models using the B-spline function. Therefore, due to the lower complexity in the (co)variance estimations, the model using Legendre polynomials represented the best option for the genetic evaluation of the Guzera lactation records. An increase of 3.6% in the accuracy of the estimated breeding values was verified when using RRM. The ranks of animals were very close whatever the RRM for the data set used to predict breeding values. Considering P305, results indicated only small to medium difference in the animals' ranking based on breeding values predicted by the conventional model or by RRM. Therefore, the sum of all the RRM-predicted breeding values along the lactation period (RRM305) can be used as a selection criterion for 305-day milk production. (c) 2014 Elsevier B.V. All rights reserved.