Figure 3
From: A neural network based model effectively predicts enhancers from clinical ATAC-seq samples
Comparisons of alternative enhancer definitions. (a) Number of OCRs mapping to enhancers based on ChromHMM, FANTOM, and p300 binding definitions. (b) Overlap of ChromHMM, FANTOM, and p300 enhancer definitions in three cell lines. All three definitions were available for GM18278, CD4+ T, whereas p300 binding data was not available in CD14+ cells. (c) Receiver Operating Characteristic curves (ROC) for enhancer predictions based on ChromHMM, FANTOM, p300, and common enhancer definitions using 5-fold cross validation. AUC stands for area under the curve (AUC) values for ROCs. Common enhancers are the enhancers annotated by all available definitions in the cognate cell type. Negative examples (“other” OCRs) were defined as OCRs that map to “other” ChromHMM states (Methods). (d) ROC curves for cross cell type predictions via models built using different enhancer definitions in GM12878 cells. Note that, models based on ChromHMM are more effective for cross cell type predictions compared to models based on common enhancers or other definitions. AUC stands for area under the curve (AUC) values for ROCs, ACC stands for accuracy.
