Abstract #M106

# M106
Genetic trends of linear type traits for validation of genomic evaluation in US Holsteins.
S. Tsuruta*1, T. J. Lawlor2, D. A. L. Lourenco1, Y. Masuda1, I. Misztal1, 1University of Georgia, Athens, GA, 2Holstein Association USA, Brattleboro, VT.

Proper modeling of genetic evaluations is necessary to obtain accurate forward predictions. Differences in genetic trends for genomic (G)PTA, traditional PTA, parent averages (PA), and daughter yield deviations (DYD) can illustrate a model’s ability to control bias due to genomic preselection and improper parameter choice. Phenotypes for 18 linear type traits and genotypes were provided by Holstein Association USA and USDA-ARS, respectively. The full data consisted of 10,067,745 records up to 2014 calving, 9,730,943 animals in pedigree, and 569,404 genotyped animals with 60K SNP. For validation of young genotyped animals who did not have phenotypes or daughters in 2010, 9,235,355 records and 105,116 genotyped animals were used to estimate genetic trends, comparing with those estimated from the full data set. The BLUP90IOD2 program was used to predict GPTA in 2010 and in 2014 with single-step genomic BLUP using the algorithm of proven and young animals. The trends were calculated separately for bulls with at least 50 daughters in 2014 and for cows with records. Assuming that GPTA in 2014 were the most accurate, GPTA in 2010 for more than half of the traits, when no parameter adjustments are made, showed some bias. Traits with directional selection, i.e., body size and udder traits, were overpredicted. Parent averages in 2014 were similar to PTA and DYD in 2014 and lower than GPTA in 2014. Traits with an intermediary optimum, such as rump angle and foot angle, showed little or no bias. Lowering the heritability slightly improved both the accuracy and predictability. Including an adjustment (weight <1.0) to the inverse of the relationship matrix of the genomic tested animals dramatically improved the predictability of the model with a slight decrease in accuracy. Future research is ongoing to fully understand how this adjustment is altering our assumptions in the basic model; e.g., how this adjustment is related to genetic parameters that could be different by generation and why some traits are not overpredicted without any adjustment.

Key Words: genomic evaluation, genetic trend, linear type trait