Reliability of genomic evaluations in Holstein-Friesians using haplotypes based on the BovineHD BeadChip

Abstract

The objectives of this study were to make subsets of high-density (HD) loci based on localized haplotype clusters, without loss of genomic information, to reduce computing time compared with the use of all HD loci and to investigate the effect on the reliability of the direct genomic value (DGV) when using this HD subset based on localized haplotype clusters in the genomic evaluation for Holstein-Friesians. The DNA was isolated from semen samples of 548 bulls (key ancestors) of the EuroGenomics Consortium, a collaboration between 4 European dairy cattle breeding organizations and scientific partners. These bulls were genotyped with the BovineHD BeadChip [~777,000 (777K) single nucleotide polymorphisms (SNP); Illumina Inc., San Diego, CA] and used to impute all 30,483 Holstein-Friesians from the BovineSNP50 BeadChip [~50,000 (50K) SNP; Illumina Inc.] to HD, using the BEAGLE software package. The final data set consisted of 30,483 animals and 603,145 SNP. For each locus, localized haplotype clusters (i.e., edges of the fitted graph model) identifications were obtained from BEAGLE. Three subsets [38,000 (38K), 116,000 (116K), and 322,000 (322K) loci] were made based on deleting obsolete loci (i.e., loci that do not give extra information compared with the neighboring loci). A fourth data set was based on 38K SNP, which is currently used for routine genomic evaluation at the Cattle Improvement Cooperative (CRV, Arnhem, the Netherlands). A validation study using the HD loci subsets based on localized haplotype clusters was performed for 9 traits (production, conformation, and functional traits). Error of imputation from 50K to HD averaged 0.78%. Three thresholds (0.17, 0.05, and 0.008%) were used for the identification of obsolete HD loci based on localized haplotype clusters to obtain a desired number of HD loci (38K, 116K, and 322K). On average, 46% (using threshold 0.008%) to 93% (using threshold 0.17%) of HD loci were eliminated. The computing time was about 9d for 38K loci, 15.5d for 116K loci, 21d for 322K loci, and 7.5d for 38K SNP. The increase in reliability of DGV compared with pedigree-based estimated breeding values for kilograms of protein was similar for 322K and 116K loci (30.7%), but was 1.5 to 2% higher compared with 38K loci and 38K SNP. Averaged over 9 traits, subset 116K loci resulted in a higher increase in reliability compared with 38K loci and 38K SNP. Eliminating obsolete loci enormously decreased the amount of data to be analyzed for genomic evaluations. The more HD loci used in a genomic evaluation, the higher the increase in reliability of DGV. It is possible to increase the reliability of DGV by 1 to 2% compared with the SNP currently used for routine genomic evaluation.