Large scale single step genomic selection in practice

Offentlig prosjektsammendrag

Automated collection of genotype and phetnoype information enabled creation of big data sets. However, these data come often with imperfections and many missing records which is causing problems in breeding values evaluation. The goal of this project is to reduce these problems and maximise the accuracy of predicted breeding values. Furthermore, we would like to prevent for potential biases that may result from the large-scale use of data on ungenotyped animals. In order to achieve this we will improve computational methods so they will be able to handle millions of animals with millions of genetic markers. We will look for chromosomal regions of importance to multiple traits and use this information to enhance prediction accuracies. We will estimate model parameters such that prediction accuracies are maximized, and biases minimized. Finally all the projects advances will be implemented into a software package that can handle large scale practical data. So far we have already decreased the bias of predictions for genotyped animals with missing pedigree information in Norwegian Red. Furthermore, we have developed new models for the estimation of breeding values for the exterior traits for Norwegian Red. The new models increased the accuracy of predicted breeding values which will enable faster genetic improvement. Finally, using machine learning methods we have developed new technique that will improve the prediction accuracy for polledness for ungenotyped animals.