摘要 : Genomic prediction is widely used to select candidatesfor breeding. Size and composition of the referencepopulation are important factors influencing predictionaccuracy. In Holstein dairy cattle, large referencepopulations are use... 展开 Genomic prediction is widely used to select candidatesfor breeding. Size and composition of the referencepopulation are important factors influencing predictionaccuracy. In Holstein dairy cattle, large referencepopulations are used, but this is difficult to achieve innumerically small breeds and for traits that are notroutinely recorded. The prediction accuracy is usuallyestimated using cross-validation, requiring the full dataset. It would be useful to have a method to predict thebenefit of multibreed reference populations that doesnot require the availability of the full data set. Ourobjective was to study the effect of the size and breedcomposition of the reference population on the accuracyof genomic prediction using genomic BLUP andBayes R. We also examined the effect of trait heritabilityand validation breed on prediction accuracy. Usingthese empirical results, we investigated the use of aformula to predict the effect of the size and compositionof the reference population on the accuracy of genomicprediction. Phenotypes were simulated in a data setcontaining real genotypes of imputed sequence variantsfor 22,752 dairy bulls and cows, including Holstein, Jersey,Red Holstein, and Australian Red cattle. Differentreference populations were constructed, varying in sizeand composition, to study within-breed, multibreed,and across-breed prediction. Phenotypes were simulatedvarying in heritability, number of chromosomes,and number of quantitative trait loci. Genomic predictionwas carried out using genomic BLUP and Bayes R.We used either the genomic relationship matrix (GRM)to estimate the number of independent chromosomalsegments and subsequently to predict accuracy, or theaccuracies obtained from single-breed reference populationsto predict the accuracies of larger or multibreedreference populations. Using the GRM overestimatedthe accuracy; this overestimation was likely due to closerelationships among some of the reference animals.Consequently, the GRM could not be used to predictthe accuracy of genomic prediction reliably. However,a method using the prediction accuracies obtained bycross-validation using a small, single-breed referencepopulation predicted the accuracy using a multibreedreference population well and slightly overestimatedthe accuracy for a larger reference population of thesame breed, but gave a reasonably close estimate of theaccuracy for a multibreed reference population. Thismethod could be useful for making decisions regardingthe size and composition of the reference population. 收起