Submitted to: Journal of Dairy Science
Publication Type: Abstract Only
Publication Acceptance Date: 3/8/2010
Publication Date: 6/24/2010
Citation: Wiggans, G.R., Van Raden, P.M. 2010. Improved reliability approximation for genomic evaluations in the United States. Journal of Dairy Science. 93(E-Suppl. 1):533(abstr. 617). Interpretive Summary:
Technical Abstract: For genomic evaluations, the time required to calculate the inverse of the coefficient matrix for the mixed-model equations increases cubically as the number of genotyped animals increases, and an approximation became necessary for estimating US evaluation reliabilities. The original approximation method used the same contribution to reliability from genomics for all animals. That method was improved by using a weighted sum of the genomic relationships of an animal with predictor animals (sumG-wt), which allowed for individual animal differences. Because calculation time for the genomic relationship matrix only increases quadratically and is routinely available, the sum of relationships of an animal with predictor animals can be obtained. Those relationships were weighted by reliability of the traditional evaluation after removing the contribution to reliability from parent average by first converting both reliabilities to daughter equivalents (DE). Reliabilities from August 2009, the last genomic evaluation for which the coefficient matrix was inverted, were decomposed to extract the genomic contribution in terms of DE calculated with an error-to-sire variance ratio of 14. Of 28,047 genotyped Holsteins, 8,353 bulls and 3,559 cows had genomic evaluations and 16,135 animals did not. Regression of DE on sumG-wt was calculated for those 3 groups. Goodness of fit was assessed by plotting predicted values against mean DE for sumG-wt groups, where groups were by 10. A straight line through the origin provided a good fit except for low sumG-wt. A floor of 30 DE was adopted to improve evaluation accuracy for animals with low sumG-wt. The slope was 0.0584 for evaluated bulls, 0.0557 for evaluated cows, and 0.0506 for animals without evaluations. The higher slope for bulls resulted in a higher reliability for the same sumG-wt. The improved approximation method increased accuracy of genomic reliabilities, particularly when comparing animals with different countries of origin and bulls with only genomic evaluations with progeny-test bulls.