215
Efficiency of Variable Selection in Genome-Wide Prediction for Traits of Different Genetic Architecture
The choice of statistical method to obtain maximum prediction accuracy in genome based-prediction is still under debate. For traits influenced by a small number of quantitative trait loci predictions should benefit from methods performing variable selection compared to methods distributing effects across the genome. However, assumptions underlying successful variable selection are frequently violated in experimental data. Based on computer simulations and experimental data from different species we investigated the breakdown behavior of different statistical methods with respect to recovering true non-zero predictors in the underlying model. The efficiency of variable selection was strongly influenced by the level of determinedness of the data, the heritability of the trait, and the extent of linkage disequilibrium in the population. Based on our results upper bounds for the number of causal mutations which can be identified by a variable selection method can be inferred.
Keywords: variable selection in genomic prediction
