Prediction of breeding values in a experimental mice population using selective genotyping

Abstract

Generally only few animals from a reference population are genotyped due to the high costs of genotyping. The aim of this research will be to evaluate which animals must be genotyped in order to get the best predictions of the genetic merit using genome-enabled prediction models. The public data from a heterogeneous population of mice will be used in this study. Public data sets of genotyped animals allows to carried out this kind of study without cost of genotyping. The data set will be divided randomly in 70% to train the models and 30% to validate them. Six selective genotyping strategies will be used: (1) 40% of the animals from the training population will be chosen at random (it will be used as a control case to compare the prediction of genetic merit with other strategies); (2) 40% of the animals with the greatest breeding values obtained by pedigree-based model via BLUP; (3) 40% of the animals with the lowest breeding values obtained by pedigree-based model via BLUP; (4) 20% of the animals with the greatest breeding values and 20% with the lowest breeding values (extremes of the population); (5) 40% of the less related animals and (6) 40% of the most related animals. These strategies will be evaluated for the body mass index (BMI, h2=0.13) and total cholesterol (CHOL, h2=0.38). The Bayesian Lasso and GBLUP will be used to predict the genetic merit of the animals. Accuracy of predictions will be assessed using Pearson's correlation between predictions and observations of genetic merits for each studied trait.