Ranking of the parental features.

<p>Parental metabolite levels (224 features in total) are used to predict the observed class labels of 54 hybrid metabolites. In each prediction model all features can be ranked according to their weights. The ranks are scaled between 0 and 1 by dividing by the total feature number. The scaled median rank distribution of a feature, <i>i.e.</i> the individual boxes in the plot, then gives an estimate regarding the importance of the absolute parental level of the respective metabolite on the heterosis pattern of all hybrid metabolites.