Genome-wide prediction (GWP) was suggested in order to overcome the shortcomings of quantitative trait loci mapping and marker-assisted selection. Genetic effects of markers are simultaneously estimated with a statistical GWP method in an estimation set consisting of genotyped and phenotyped individuals. The objective of the present study was the development of novel ridge regression methods that improve existing GWP methods with respect to accuracy of predicted genotypic values, accuracy of marker effect estimates and computational efficiency. For this purpose, their properties were compared in simulated data and data sets from applied plant breeding programs of maize, wheat and sugar beet.Prediction of test cross performance was investigated for the first time with an independent validation of a data set originating from two subsequent breeding cycles of an applied sugar beet breeding program. It was demonstrated that genetic effects which were estimated in a certain cycle of a breeding program can be used for prediction of genotypic values in the subsequent breeding cycle, if the trait under consideration has a high heritability, as for example sugar content. For the prediction of genotypic values, ridge regression employing preliminary estimates of the heritability (RIR) was the fastest GWP method among those employing homoscedastic marker variances.Accurate estimation of the true genetic effects for each marker is an important criterion for heteroscedastic GWP methods, if they are used for the identification of functional genes for gene introgression or the prediction of the performance of crosses. A modification of the expectation-maximization algorithm that yields heteroscedastic marker variances (RMLV) and ridge regression with weighing factors according to analysis of variance components (RRWA) provide alternative solutions to the computationally demanding Bayesian methods. RRWA outperformed all of the investigated GWP methods employing heteroscedastic marker variances in terms of computational efficiency. Most accurate marker effects in a simulated data set were estimated using RMLV, especially in situations with long LD stretches along the chromosomes and high marker densities, which often occur in plant breeding programs.It can be concluded that the proposed novel ridge regression methods are promising for providing accurate predictions of genotypic values, accurate marker effect estimates and computational efficiency.
Verknüpfung zu Publikationen oder weiteren Datensätzen
