Fig. 3

Final decision tree developed on qPCR validation dataset. a The structure of best performing expression-based decision tree for differentiation of samples; b The predictive abilities of the model by ROC analysis. Class probability estimates based on observed relative class frequencies at the leaf nodes were used for the development of ROC curves. AUC area under the ROC curve, DNET dysembryoplastic neuroepithelial tumors, GG gangliogliomas, PA pilocytic astrocytomas, ROC receiver operating characteristic curve