Fig. 2: Properties of the CNEP score.

a The graph shows the cumulative distribution of the CNEP score genome-wide (Genome), in PhastCons constrained non-exonic (CNE) bases, and bases that are not in PhastCons-constrained elements and also not in exons (notCNE). b A scatter plot with each point corresponding to one feature that CNEP uses. The x-axis shows the average CNEP score in bases that have the feature present, while the y-axis shows the expected CNEP score based on the feature’s overlap with constrained non-exonic bases. Only 48,364 features that cover at least 200 kb are shown. The full set of values can be found in Supplementary Data 2. The diagonal line is the y = x line. The vertical line corresponds to the genome-wide average CNEP score. The horizontal line corresponds to the genome-wide expected average CNEP score. c A plot showing the average fraction of the 350 Roadmap DNase I experiments in which the base overlaps a called peak for each CNEP score value, rounded to the nearest 0.001, covering at least 1000 bases. In total, there was 1000 such values. d A plot showing the average fraction of bases annotated across the 127 epigenomes to each of 14-groups defined based on 25 ChromHMM chromatin states previously assigned the same color⁴⁰ for each CNEP score value, rounded to the nearest 0.001. A color legend with the state mnemonics from ref. ⁴⁰ is displayed at the bottom of the panel. e A plot of the ROC curve for the CNEP score predicting PhastCons non-exonic bases. Also shown is the performance of individual features and several baseline or existing scores (see “Methods” section). Area under the curve values are shown in parentheses. f A similar plot as e except for precision-recall as opposed to ROC curves. ROC and precision-recall curves for other constrained element sets can be found in Supplementary Fig. 5. Source data are provided as a Source Data file.