Fig. 3

ChromScoreHMM emission parameters and enrichments. A Emission parameters of a ChromScoreHMM model learned based on combinatorial and spatial patterns of top scoring predictions of each expert (top 2% of predictions, Methods). Each row of the heatmap corresponds to a ChromScoreHMM state (states 1–15, color legend on left margin) and each column a different input expert model. Emission parameter values correspond to the probability in that state of observing a top scoring prediction for that expert model. B Overlap fold enrichments for (1) sequence and gene based annotations: CpG islands [40], exons, gene bodies, and transcription start sites from RefSeq [41], CTCF motifs from HOMER [42], (2) evolutionary conservation related annotations: GERP++ [43] and PhastCons 100 vertebrates conserved elements [44], (3) ERV1, LINE and LTR repeat elements from RepeatMasker [45], (4) ChromHMM annotations, 25-state model [13, 39]. Top row (“Genome % (Annotations)”): Percentage of the genome covered by each annotation type, calculated by dividing the number of genomic base pairs covered by a given annotation by the total size of the genome. As some annotations can overlap each other, the sum of these percentages exceeds 100%. C Percentage of the genome assigned to each ChromScoreHMM state. See Additional file 1: Fig. S8 and Additional file 1: Fig. S9 for additional enrichments. Red shading: emission parameters, blue shading: fold enrichments, black shading: genome percentages. Enrichments and percentages are medians across cell types