Multi-scale annotations of chromatin states in 127 human cell-types

Abstract

Genome-wide profiling of chromatin states has been widely used to characterize the biological function of non-coding genomic sequences in a cell-type specific manner. However, the systematic, comprehensive annotations of chromatin states from experimental data are challenging and require not just extensive biological knowledge but also sophisticated computational modeling. Previously we developed a hierarchical hidden Markov model, named diHMM, to systematically annotate chromatin states at multiple scales based on the combination of histone mark and chromatin regulator binding profiles. Here, we have improved the method by optimizing computational efficiency and using an ensemble-clustering approach to achieve a unified annotation by integrating information from cell-type-specific models. We then applied this improved method to generate a unified multi-scale chromatin state map in 127 human cell types, based on public data generated by the Epigenome Roadmap and ENCODE consortia. We found cell types with similar origin are typically associated with similar chromatin states, but cultured cell lines have distinct structures than primary cells. The contribution of enhancer elements to gene regulation is mediated by the broader context of domain-state organization. Distinct domain-state patterns are associated with various 3D chromatin structures. As such, we have demonstrated the utility of the multi-scale chromatin state map in characterizing the biological function of the human genome.