Experiences with the Illumina high density bovine BeadChip

Abstract

The effect of increasing SNP density on the accuracy and inflation of genomic predictions of protein yield was investigated. Three SNP densities were studied. One was based on the 50K chip (38,296 SNPs). The two others were based on the high-density (HD) chips; HD1 (692,598 SNPs) and a subset, HD2 (329,329 SNPs), obtained by reducing the multicollinearity of the HD1 SNPs. The data set consisted of 4211 Holstein Friesian (HF), Jersey (J) and HFxJ sires genotyped with the 50K chip. The HD1 data was obtained through either genotyping or imputation. The test animals (N=605) were the two youngest cohorts of the progeny-tested sires. The genomic breeding values (GBVs) of the test animals were predicted using Bayesian and nonparametric methods. All models included the polygenic effect.

A higher SNP density was found to slightly improve the accuracy of prediction when Bayesian ridge regression was used to train on one breed and predict on a different breed. When using an across-breed analysis, the increase in SNP density did not increase the accuracy of predicting the GBVs. Furthermore, the genomic predictions were inflated compared to the model that included only the polygenic effect, with predictions obtained using the HD genotypes having greater inflation than those obtained from the 50K genotypes. The training data set may not have been large enough to take full advantage of the HD genotypes.