Figure 2.
Validation of non-linear classification models fit of the random forest algorithm. Differentially expressed genes or pathways in the discovery set are used to train the model, whose performance is evaluated in the validation set. The performance is evaluated via the area under (AU) receiver operating characteristic (ROC) or precision-recall gain (PRG) curves. Black lines: curves for models trained on transcriptome data; grey lines: curves for models trained on annotated pathways.