BayesAS and GBLUP models for genomic selection in Mink T. Villumsen1, G. Su 1, Z. Cai 1, B. Guldbrandtsen 1, T. Asp 1, G. Sahana 1, M.S. Lund 1 1 Department of Molecular Biology and Genetics, Aarhus University (DK). Conclusions: Multiple trait models generally didn’t perform better than single trait models for our mink data BayesAS models allowing heterogeneous (co)variance over the genome were superior to GBLUP models Questions: Does a multiple trait model perform better than a than single trait model? Is a model that accounts for heterogeneous (co)variance over the genome (BayesAS)superior to a GBLUP model for our mink data? Results: Data: Foulum brown line 28,336 SNP-markers in 391 scaffolds 2013 Accuracy of prediction in the 5-fold cross validation for 2014 200 Males Geno- & Phenotypes 800 Females Geno- & Phenotypes   Trait No of Animals ST-GBLUP MT-GBLUP ST-BayesAS MT-BayesAS Live grad. Weight1 752 0.490 0.5001 0.516 0.5231 Quality 0.685 0.694 0.695 0.706 Density 0.393 0.462 0.450 0.459 Silky 0.820 0.818 0.844 0.843 Dried skins P_length 639 0.480 0.489 0.618 0.624 P_quality 665 0.227 0.234 0.270 0.258 P_density 0.299 0.302 0.357 0.362 P_silky 0.138 0.135 0.197 0.194 6000 Progeny Live- and Peltgrading 2014 200 Males Geno- & Phenotypes 400 Females Geno- & Phenotypes 6000 Progeny Live- and Peltgrading 840 Genotyped Models: MT-GBLUP: MT-BayesAS: Hierarchial model, handle correlation between SNP effects: Not a clear tendency for MT-models to be better than ST-models BayesAS superior to GBLUP models BayesAS GEBV increased more for pelt traits than live grading traits All traits analyzed with weight in MT-analyses. Results for weight: analyzed with P_length AU AARHUS UNIVERSITY DEPARTMENT OF MOLECULAR BIOLOGY AND GENETICS