Using herd-averages of feed efficiency as training data for genomic selection

Abstract

Genomic selection on feed efficiency traits in dairy cows can save more feed costs and result in more sustainable dairy industry. Here we estimate the accuracy of genomic selection regarding dry matter intake in dairy cattle on individuals’ and herd levels for training and validation animals. The training population consists of 27856 cows from 833 herds. The validation population was 1104 cows from 11 herds. The number of single nucleotide polymorphisms (SNPs) used for genomic prediction was 41227. The simulated heritability for dry matter intake was 0.25. The accuracy of genomic selection for the training animals was 0.799, and for the validation animals was 0.748. The herd-wise average genotype of the training animals from 833 herds of the 41227 SNP-chip genotypes and the average phenotype of these 833 herds were used to estimate the SNP effects. The accuracy of genomic selection of 833 herds from training animals and GEBV estimated by average genotype per herd from training animals was 0.345 and 0.495. Furthermore, using these 41227 estimated SNPs effects from the herd-average genotypes and phenotypes for the prediction of GEBV of the validation animals resulted in ~0 correlations between TBV and GEBV of the validation animals. It was concluded that using a large number of individual phenotypic records will achieve high accuracy of genomic selection in dry matter intake for dairy cows. Genomic selection with herd-wise averaged genotypes and phenotypes as training data did not yield prediction accuracy for validation animals in this study because herd-averaged genotypes resulted in decreased variance in genotypes and phenotypes, which leads to the less precise estimates of the SNP effect to predict the GEBV of validation animals and reduces the accuracy of the estimates of the SNP effects.