Figure 9

Mean Decrease Accuracy (%IncMSE) and Mean Decrease Gini (IncNodePurity) (sorted decreasingly from top to bottom) of attributes as assigned by the random forest. The abbreviations of the descriptors can be found in table 1 of additional file 1. For the mean decrease in accuracy the most relevant descriptors either relate to the proton or the carbon it is connected to or to atoms close to this. The most important types of descriptors are hybridisation, electronegativity, distances and whether the proton is joined to a conjugated π system.