Fig. 4

Chromatin factor binding predictions in validation cell types and Roadmap datasets. a Number of genomic bins that chromatin factor is predicted to bind (left) and MCC (right) as a function of posterior probability cutoff for SRF (top) and CTCF (bottom). This relationship is shown for H1-hESC (turquoise), 2 validation cell types for SRF (blue), and 6 validation cell types for CTCF (blue). Each curve represents predictions for one of the 4 chromosomes (chr5, chr10, chr15, and chr20). Left panels also show how many genomic bins are predicted to bind the chromatin factor in 18 Roadmap datasets (red). Vertical red dashed line: posterior probability cutoff which maximized MCC of the chromatin factor in H1-hESC. Horizontal red dashed lines: number of genomic bins with chromatin factor binding in validation cell types. b Boxplot of various performance measures when using the best cutoff for each dataset (red) and the optimal cutoff in H1-hESC (turquoise). c UCSC Genome Browser display of a 4000 bp region on chromosome 20 using the Virtual ChIP-seq track hub (https://virchip.hoffmanlab.org). The track hub has a supertrack for each chromatin factor. Each supertrack contains 34 tracks: one track specifying genomic bins bound by that chromatin factor in Cistrome and ENCODE, and one track for each of the 33 Roadmap cell types with predictions for that chromatin factor. This example shows parts of the track hub related to CTCF, including a track with experimental results in Cistrome DB and ENCODE with 7 out of 144 cell types enabled, and Virtual ChIP-seq predictions in left lung, adrenal gland, B-cell, and T-cell. The height of predictions indicates the number of overlapping genomic bins predicted to bind the chromatin factor, ranging between 0 and 4. Between are MACS2 narrow peak calls for CTCF in NHLFs from ENCODE (ENCFF510CUI). Blue: peaks; orange: peak summits