Fig. 4

Identification of optimal minimal sets of epigenetic marks for predicting functional states of CRMs in human cell/tissue types. A Upset plot showing the mean unique genome positions of the peaks of each mark and the mean intersections of genome positions covered by the peaks of the four marks (upper and middle panels), and boxplot of AUROCs of the 15 LR models using the 15 different combinations of the four marks in the 67 human cell/tissue types. Models 13, 14, and 15 have similar AUROCs, each is significantly higher than those of the other 12 models, **p<0.001 (Wilcoxon signed-rank test). B ROC curves of model 15 using the four marks in the human cell/tissue types. Each gray curve is the ROC curve for a cell/tissue type, and the red one is the mean ROC curve of the 67 cell/tissue types. C Boxplot of weights (coefficients) of the four marks in model 15. **p<0.001 (Wilcoxon signed-rank test), the weights of CA are significantly higher than those of the other three marks