Comparison of genomic selection approaches in Brown Swiss within Intergenomics

Abstract

The European Brown Swiss federation, in collaboration with Interbull, funded and managed a project named Intergenomics. The goal of this project is to perform genomic evaluations of sires based on a joint analysis of all the genotypes collected around Europe. To date, six countries are involved in Intergenomics and according to the country, between 3 and 15 traits are available. In this study, we propose to compare a panel of 4 genomic selection approaches to the pedigree-based BLUP (Best Linear Unbiased Predictor). Among these 4 methodologies, performances of the genomic BLUP (GBLUP) were compared to 2 bayesian approaches (Bayesian LASSO and Bayes Cπ) and a variable selection approach (Elastic Net or EN). Except the GBLUP, the other genomic selection approaches deal with the p>>n problem (number of Single Nucleotide Polymorphism or SNP (p) is much higher than the number of bulls (n)).

We compare the correlations between observed and predicted deregressed proofs for the different traits, the different country scales and the different methods. Compared to the pedigree-based BLUP, genomic selection approaches allow a gain in correlation between 6.5 and 20.9%. Bayesian LASSO, Bayes Cπ and EN give the best results with a gain of correlation around 3% compared to a GBLUP. The slope of regression is also lowest with these three methods than with the pedigree-based BLUP and the GBLUP. Consequently, over the different country scale, the mean number of traits which validate the interbull test (slope of regression between 0.8 and 1.2) is lowest for the pedigree-based BLUP (6.4 traits in average) than for the Bayesian LASSO, Bayes Cπ and EN (between 7.8 and 8 traits in average).