Skip to main content
ARS Home » Research » Publications at this Location » Publication #278338

Title: Methods to include foreign information in national evaluations

item Vanraden, Paul
item Tooker, Melvin

Submitted to: Journal of Dairy Science
Publication Type: Abstract Only
Publication Acceptance Date: 3/5/2012
Publication Date: 6/28/2012
Citation: Van Raden, P.M., Tooker, M.E. 2012. Methods to include foreign information in national evaluations. Journal of Dairy Science. (Suppl. 2):446–447(abstr. 449). 2012.

Interpretive Summary:

Technical Abstract: Genomic evaluations (GEBV) with higher reliability often result from including genotypes and phenotypes from foreign bulls in the reference population. Multi-step methods evaluate domestic phenotypes first using only pedigree relationships (EBV), then add foreign data available from multi-trait across country evaluations (MACE), then compute GEBV for genotyped animals, and finally propagate information from GEBV to EBV of non-genotyped relatives. An alternative is to include domestic and foreign phenotypes together so that GEBV and EBV for all animals can be computed in a single step. The MACE EBV could be treated as a correlated trait, but previous research indicates that including these as the same trait with their lower reliability REL is sufficient. To include foreign data, the bull’s deregressed proof (DRP) was obtained from the MACE EBV as: DRP = PA + (EBV – PA)/REL, where PA is the parent average from MACE. For bulls with both domestic and foreign daughters, domestic EBV was used instead of PA to compute DRP, and domestic daughter equivalents (DE) were subtracted from the total. Remaining DE were added to diagonals of the mixed model equations and were used to compute REL. This strategy included 1 extra record per bull and differed from previous methods that included 1 record for each foreign daughter. For multi-trait models, diagonal matrix D contained the DE for each trait of a bull. The vector of DRP was pre-multiplied by D^.5 T^-1 D^.5, where T is the genetic covariance matrix among traits, and D^.5 T^-1 D^.5 was added to the mixed model equations. A mean for the DRP was included in the model because the base is not fixed during iteration, only after convergence. The methods were tested using national Holstein data for 25 million cows, MACE data for 88,000 bulls, and a pedigree file of 52 million animals. For bulls with only foreign daughters, correlations between MACE EBV and national EBV after including the foreign data were 0.991 to 0.994 for yield traits, 0.986 for somatic cell score, 0.973 for single-trait productive life, and 0.974 for daughter pregnancy rate. This simple approach is reasonably accurate for including foreign data in national evaluations.