Fig. 2

Chromatin state analysis using ChromHMM. a Illustration of peak consistency between raw and normalized data for nine histone marks that were used for chromatin state analysis for nine different cell lines, as indicated. X and Y-axis of the plot are the percentage of peaks overlapping between normalized and raw data, respectively. The least overlapping rate was observed for the H3K27ac profile of H1 cells, where all the peaks from raw data (100%) were retained post normalization but only 25% of peaks from normalized data overlapped with raw data peaks showing that additional peaks were identified post normalization. As for the H3K27ac profile of H1 cells, the poor overlap between peaks predicted from raw and normalized profiles was generally due to either poor quality and/or low coverage. b Emission parameters of ChromHMM describing chromatin state differences between raw and normalized peaks. Though the predicted chromatin states were conserved, three significant differences in enrichment levels are highlighted as red-framed boxes. c An example region illustrating the change after normalization of chromatin state 14 in Fig. 2b, where H3K27ac peaks become prominent after normalization. d Stacked bar chart indicating the percentage of chromatin state annotations per bin that changed upon normalization. While the GM12878, NHEK and NHLF datasets show few changes after normalization, the other datasets show more than 5% changed bin annotations. e Illustration of change in chromatin state annotation for the MYO7A locus using the same dataset processed with ChromHMM; note that the MYO7A promoter was annotated ‘active’ from the raw data and changed to ‘poised’ post normalization, which correlates perfectly with the absence of gene expression [Encode data: ENCSR962TBJ]