Abstract #172

# 172
Implementing SNP-level multiple-trait across country genomic evaluation without genotype sharing.
Breno Fragomeni*1, Daniela Lourenco1, Yutaka Masuda1, Ignacy Misztal1, 1The University of Georgia, Athens, GA.

There is a growing interest of Interbull in releasing a multiple across country genomic evaluation. However, most countries are not able to provide genotypes, and an alternative methodology is required. One strategy called SNP MACE posits a multiple-trait SNP BLUP based on left- and right-hand sides of national SNP BLUP. However, different countries use different sets of SNPs and multiple-trait computations with SNP may be difficult. We propose an alternative model based on reconstructing phenotypes for an independent genotyped population. Each country would submit only SNP effects, the number of reference animals, and average reliabilities of GEBV. This information can be used to create a pseudo-population with pseudo-observations. The combined data can be analyzed by multi-trait GBLUP. Conversion of GEBV would provide SNP effects in scale of every country. Simulations included 30k animals resembling the US Holstein population, with effective population size of 120. Chromosome number and size mimicked the cattle genome. The population was then divided in 3: 2 countries and 1 test population with 10k genotyped animals in each, and a different trait was assigned to each country. For the genotyped animals in the 2 countries, DYD were generated with an average reliability of 0.8. SNP effects were calculated with GBLUP in each one of the 2 countries. With SNP effects from the 2 countries, phenotypes were reconstructed for the test population. A bivariate GBLUP was then fitted, and GEBV/DGV were calculate for the test population for both countries. Accuracies were calculated for the validation population on the scale of 2 countries. When SNP effects of one country were used, the realized accuracy was 0.94 for the same population and 0.69 for the second country. When SNP effects of both countries were used, the accuracy for any country was 0.95. With the use of the APY algorithm, the procedure is computationally viable for any population size and any number of countries. An important issue is creation of pseudo-population that holds the same genomic information as the national population.

Key Words: SNP-MACE, genomic MACE, SNP effect, Interbull