|De Vries, A|
Submitted to: Journal of Dairy Science
Publication Type: Abstract Only
Publication Acceptance Date: 2/23/2013
Publication Date: 7/8/2013
Citation: De Vries, A., Cole, J.B., Galligan, D.T. 2013. Regression metamodels of an optimal genomic testing strategy in dairy cattle when selection intensity is low. Journal of Dairy Science. 96(E-Suppl. 1):74 (abstr. T207). Interpretive Summary:
Technical Abstract: Genomic testing of dairy cattle increases reliability and can be used to select animals with superior genetic merit. Genomic testing is not free and not all candidates for selection should necessarily be tested. One common algorithm used to compare alternative decisions is time-consuming and not easily applicable in practice. Therefore, the objective of this study was to develop a regression metamodel that predicts increases in estimated breeding value (EBV) of net merit (NM$) in selected animals based on the reliability of pre-ranking of all animals, reliability of the genomic test, proportion of pre-ranked animals that are genomically tested, and selection intensity. First, the increase in EBV NM$ in selected animals (>50% of the population) was calculated using Monte Carlo methods when all animals were pre-ranked for EBV NM$ with reliabilities varying from 0 to 100% in increments of 10 percentage points (PP). After pre-ranking all animals, the genomic test was applied to all ranges of pre-ranked animals in 10 PP increments (n=36,300 scenarios). Selection was applied after the second ranking and the predicted gain in EBV NM$ recorded. For example, predicted gain in EBV NM$ with 20% pre-rank reliability, testing the 60 to 90 percentiles of the pre-ranked animals, 60% genomic test reliability, and 90% selection intensity was $80. Second, the SAS procedure glmselect was used to develop regression metamodels that predict gain in EBV NM$ given 30 variables constructed from reliabilities, ranges of genomically tested animals, selection intensity and their logs, squares and reciprocals. Models constrained to 5, 10, 20, or 40 variables including 2-way interactions had RMSE of $14.90, $10.98, $6.47 and $5.11, respectively. The R-squared ranged from 94.4% to 99.4%. The same 4 models including 4-way interactions had RMSE of $12.20, $6.61, $3.62, and $2.45. The R-squared ranged from 96.3% to 99.9%. In conclusion, the larger metamodels accurately predicted gain in EBV NM$ and can easily be implemented in decision support aids. The cost of genomic testing may be added to find the optimal range of pre-ranked animals that should be genomically tested.