Fig. 4From: Genome-wide prediction of cis-regulatory regions using supervised deep learning methodsFeature importance and classification performance in the 3-class (A-E versus A-P versus BG) scenario. a Feature importance discovered by randomized DFS (RDFS) and random forest (RF) on GM12878. The random forest’s feature importance scores were normalized to [0,1] for better comparison with randomized DFS. b auPRC versus the number of features incorporated into the RDFS and RF. The annotated points indicate where a line with slope 0.5 intersects a fitted curveBack to article page