Fig. 1

Overview of the ChromActivity framework. A Flowchart of the ChromActivity framework. ChromActivity takes as input regulatory activity labels from targeted genomic regions from k different functional characterization datasets (stacked white blocks, upper left). Using features based on chromatin mark signals, peak calls, and chromatin state annotations for the targeted regions (red block, lower left) which it preprocesses (purple block, lower left), it trains a separate classifier (“expert”) for each functional characterization dataset. Each expert provides a predicted genomewide regulatory activity score track specific to a functional characterization dataset (stacked blue blocks). ChromActivity then uses the score tracks to generate two complementary outputs reflecting predictions of regulatory activity for each cell type (yellow blocks, right): (i) ChromScoreHMM annotations, which are annotations of the genome into states generated by integrating combinatorial and spatial patterns in the expert prediction score tracks using ChromHMM and (ii) ChromScore tracks, which are continuous genomewide regulatory activity score tracks based on the mean individual expert scores at each 25-bp interval. B Visualization of regulatory activity score tracks for each expert, ChromHMM chromatin state annotations (25-state imputed model), the ChromScore track, and ChromScoreHMM annotations in HepG2 for genomic interval chr1:6,000,000–6,100,000 (hg19). ChromScoreHMM and ChromHMM color legends are shown in Additional file 1: Fig. S1