Use of Repeated Group Measurements with Drop Out Animals for Variance Component Estimation and Genetic Evaluation: A Simulation Study

The efficiency of feed utilization plays an important role in animal breeding. However, measuring feed intake (FI) is costly on an individual basis under practical conditions. Using group measurements to model FI could be practically feasible and cost-effective. The objectives of this study were to develop a random regression model based on repeated group measurements with consideration of missing phenotypes caused by drop out animals. Focus is on variance components (VC) estimation and genetic evaluation, and to investigate the effect of group composition on VC estimation and genetic evaluation using simulated datasets. Data were simulated based on individual FI in a pig population. Each individual had measurement on FI at 6 different time points, reflecting 6 different weeks during the test period. The simulated phenotypes consisted of additive genetic, permanent environment, and random residual effects. Additive genetic and permanent environmental effects were both simulated and modeled by first order Legendre polynomials. Three grouping scenarios based on genetic relationships among the group members were investigated: (1) medium within and across pen genetic relationship; (2) high within group relationship; (3) low within group relationship. To investigate the effect of the drop out animals during test period, a proportion (15%) of animals with individual phenotypes was set as the drop out animals, and two drop out scenarios within each grouping scenario were assessed: (1) animals were randomly dropped out; (2) animals with lower phenotypes were dropped out based on the ranking at each time point. The results show that using group measurements yielded similar VCs estimates but with larger SDs compared with the corresponding scenario of using individual measurements. Compared to scenarios without drop out, similar VC estimates were observed when animals were dropped out randomly, whereas reduced VC estimates were observed when animals were dropped out by the ranking of phenotypes. Different grouping scenarios produced similar VC estimates. Compared to scenarios without drop out, there were no loss of accuracies of genetic evaluation for drop out scenarios. However, dropping out animals by the ranking of phenotypes produced larger bias of estimated breeding values compared to the scenario without dropped out animals and scenario of dropping out animals by random. In conclusion, with an optimized group structure, the developed model can properly handle group measurements with drop out animals, and can achieve comparable accuracy of genetic evaluation for traits measured at the group level.