Comparative annotation of functional regions in the human genome using epigenomic data

Epigenetic regulation is dynamic and cell-type de-pendent. The recently available epigenomic data in multiple cell types provide an unprecedented oppor-tunity for a comparative study of epigenetic land-scape. We developed a machine-learning method called ChroModule to annotate the epigenetic states in eight ENCyclopedia Of DNA Elements cell types. The trained model successfully captured the characteristic histone-modification patterns asso-ciated with regulatory elements, such as promoters and enhancers, and showed superior performance on identifying enhancers compared with the state-of-art methods. In addition, given the fixed number of epigenetic states in the model, ChroModule allows straightforward illustration of epigenetic variability in multiple cell types. Using this feature, we found that invariable and variable epigenetic states across cell types correspond to housekeeping functions and stimulus response, re-spectively. Especially, we observed that enhancers, but not the other regulatory elements, dictate cell specificity, as similar cell types share common en-hancers, and cell-type–specific enhancers are often bound by transcription factors playing critical roles in that cell type. More interestingly, we found some genomic regions are dormant in cell type but primed to become active in other cell types. These obser-vations highlight the usefulness of ChroModule in comparative analysis and interpretation of multiple epigenomes