Application of genomic evaluation to dairy cattle in New Zealand.

A hybrid method of estimating genomically enhanced breeding values (GEBV) that uses features of a single-step approach and traditional national estimated breeding value (EBV), based on pedigree, as the starting point was developed and assessed. The method uses deregressed EBV from the traditional EBV estimation procedures and a pedigree matrix that is augmented by a genomic relationship matrix based on a Euclidean distance (EDM) in a Gaussian kernel. Genotypes were obtained using the Illumina BovineSNP50 BeadChip (Illumina Inc., San Diego, CA). The pedigree information is restricted to animals with genotypes, as well as all of their ancestors, resulting in a system of equations that is considerably smaller than that of the national evaluation. Genome-wide association studies were used to identify regions of the genome that had high association with the traits in the New Zealand national breeding objective. The EDM were created for each selected region. The amount of variance accounted for by each region was estimated, and the variances, relative to the EDM excluding these areas, were used to create trait-specific weighted EDM that were used for the evaluations. The evaluation is done on the restricted data set and is a reversal of the deregression procedure using the augmented pedigree-based relationship matrix. The method was tested using data on 7,204 progeny-tested sires, in which the daughter records for the 3 youngest cohorts (i.e., validation data; n=1,410) were excluded. Evaluations were done for the 7 traits (3 production, 4 nonproduction) in the New Zealand national breeding objective. Regression of the traditional EBV based on daughter information on the GEBV for validation sires had an average slope of 0.99 (range 0.93 to 1.06); the average correlation between the EBV was 0.79 (range 0.56 to 0.84). The average slope and correlation for the previously used blended GEBV were 0.64 and 0.61, respectively. The method was applied to all genotyped animals that included approximately 13,000 genotyped males and 47,000 genotyped females and their ancestors. The correlation between the traditional EBV and GEBV of the proven bulls was 0.99; the correlations obtained from the previously used blended GEBV were 0.88 (no inflation adjustment) and 0.94 (following inflation adjustment). The values for the unproven bulls were 0.95 (parent average EBV and GEBV), 0.56 (blended GEBV, no inflation adjustment), and 0.88 (blended GEBV, following inflation adjustment). The use of trait-specific weighted EDM provided a 1 to 2% improvement in the accuracy of the genomic evaluations of the production traits and no improvement in the nonproduction traits. The hybrid method of estimating GEBV was found to be considerably less biased and generally more accurate than the previously used blended genomic breeding value.