Submitted to: Interbull Annual Meeting Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 2/2/2012
Publication Date: 2/2/2012
Citation: Van Raden, P.M. 2012. Avoiding bias from genomic pre-selection in converting daughter information across countries. Interbull Annual Meeting Proceedings. Interbull Bull. 45, 5 pp.
Interpretive Summary: Breeders have exchanged and converted genetic evaluations of bulls across countries for decades, but traditional evaluations may become biased by pre-selection on genotype. Bulls that were pre-selected on genotype now have daughter records, and new methods are needed to include both foreign and genomic information in 1-step or multi-step evaluations. This study compared U.S. Jersey traditional evaluations (EBV), genomic evaluations (GEBV), daughter yield deviations (DYD) and genomic DYD (DYDg). Correlations between the 1-step and multi-step method GEBV for young U.S. bulls were fairly high (0.962) as compared to the correlations between GEBV and parent averages (0.853 to 0.869). Correlations between DYDg and DYD were very high for U.S. bulls (>0.9993) regardless of the inclusion of foreign or genomic information. These results indicate that DYDg can account for pre-selection of bulls, and exchanging DYDg across countries could eliminate the need to partition genomic from phenotypic information when used in across country evaluations which will help increase accuracy of evaluations that aid breeders in identifying superior animals.
Technical Abstract: Methods to include both foreign and genomic information in single-step or multi-step evaluations were developed and compared using the U.S. national Jersey database. Breeders have exchanged and converted genetic evaluations of bulls across countries for decades, but traditional evaluations may become biased by pre-selection on genotype. When foreign and genomic data were added to the equations, daughter yield deviations computed from only domestic daughter records were very stable. Those could be exchanged internationally, thereby avoiding the difficulty of deregressing genomic evaluations. A final step in the multi-step method simply inserted the genomic evaluations and held them constant during iteration instead of adjusting the data vector and equations. For genotyped young bulls, multi-step evaluations were correlated by .962 to single-step evaluations computed with an algorithm that did not require inverting the genomic relationship matrix. Accuracy was similar but regressions were closer to expectation for the single-step evaluations.