Abstract:

Extensive genetic information enables the identification of genome regions associated with trait expression but this may lead to high model dimensions and the occurrence of dependencies among predictor variables. A grouped penalised regression approach provides a framework to account for these dependencies adequately. Previous studies showed that the overall performance of genomic evaluations improved if genomic markers were grouped according to their strength of interdependence. Here, we investigate different options for grouping taking linkage and linkage disequilibrium into account and additionally considering adjacency between markers. As the extent of dependence is influenced by the family structure in a livestock population, we investigated an empirical and a theoretical measure thereof. Options have been validated with simulated data. Generally, the more groups were determined, the better the performance of the grouped penalisation approach.