Evaluating the impact of uncertainty in estimated breeding values on optimal contribution selection

Many livestock breeding programmes have a closed nucleus and therefore have to manage the conversion of genetic variation into genetic gain with care. Consequently, these breeding programmes use some methods for managing genetic variation. One such method is the optimal contribution selection (OCS), which aims to maximise genetic gain for a given loss in genetic variation. The accuracy of this optimisation depends on the accuracy of its input parameters; the estimated breeding values (EBV) and the relationship matrix. In OCS, we typically assume that EBV are true values, that is, that they have no associated uncertainty, which is seldom the case. The aim of this study was to evaluate how uncertainty in EBV impacts optimised contributions and success of OCS. To this end, we have stochastically modelled a small breeding programme and obtained 1,000 posterior samples of EBV from either the pedigree (BLUP) or genomic (GBLUP) evaluations. We then ran OCS for each EBV sample (probabilistic OCS), for the posterior mean of EBV (naïve OCS), as well as for the true breeding values (true OCS). We finally evaluated the distribution of optimised contributions for the selected parents and of genetic mean and genetic variation of their progeny. We compared scenarios with OCS that aimed for an effective population size of 50 and 100. Results have shown considerable underestimation of optimised contributions for naïve OCS compared to probabilistic OCS, which were closer to the true OCS. Future work includes the use of robust optimisation that can work with a full distribution of estimated breeding values, thereby enabling probabilistic OCS for routine use.