|
An issue for implementation of single step genomic evaluations is how to weight genomic and pedigree relationships in modelling genetic co-variance. A weighting parameter lambda ranging between 0 and 1 can be used in the statistical model, with higher values corresponding to greater weighting of genomic information. We investigated appropriate values of lambda for a range of carcass traits in terminal sire sheep breeds, using the accuracy and bias of genomic prediction of breeding values as criteria. The accuracy generally increased with lambda, although the "optimal" value of lambda at the maximum accuracy varied widely, covering almost the entire range of possible values across traits. Accuracy typically approached an asymptote towards the optimal lambda, so a wide range of values could be used with minimal loss of prediction accuracy. The bias in Estimated Breeding Values (EBVs) increased with lambda, such that EBVs over-predicted phenotypic performance at high values of lambda. |
|