Chromatin-state discovery and genome annotation with ChromHMM

Abstract

SUMMARY Non-coding DNA regions play central roles in human biology, evolution, and disease. ChromHMM helps annotate the non-coding genome using epigenomic information across one or multiple cell types. It combines multiple genome-wide epigenomic maps, and uses combinatorial and spatial mark patterns to infer a complete annotation for each cell type. ChromHMM learns chromatin state signatures using a multivariate hidden Markov model that explicitly models the combinatorial presence or absence of each mark. ChromHMM uses these signatures to generate a genome-wide annotation for each cell type, by calculating the most probable state in each genomic segment. ChromHMM provides automated enrichment analysis of the resulting annotations to facilitate the functional interpretations of each chromatin state. ChromHMM is distinguished by its modeling emphasis on combinations of marks, its tight integration with downstream functional enrichment analyses, its speed, and its ease of use. Chromatin states are learned, annotations produced, and enrichments computed within one day.

Integrating the Epigenome to Identify Drivers of Hepatocellular Carcinoma

“…Ingenuity Pathway Analysis (IPA; Qiagen) is used for gene ontology and comparative analyses. Basic summarization, visualization, and statistical analysis are performed in R. Phyloepigenetic trees are constructed using the R package “ape.” Chromatin Hidden Markov modeling (ChromHMM) is used to construct histone modification state models . Browser views were obtained from the University of California Santa Cruz Genome Browser.…”

“…Chromatin Hidden Markov modeling (ChromHMM) is used to construct histone modification state models. (16) Browser views were obtained from the University of California Santa Cruz Genome Browser. Additional description is provided in the Supporting Material.…”

Evaluating the informativeness of deep learning annotations for human complex diseases

“…In our analysis of variant-level deep learning annotations across 41 traits, we analyzed 25 non-tissue-specific deep learning annotations: 8 DeepSEA annotations 13,15 previously trained to predict 4 tissue-specific chromatin marks (DNase, H3K27ac, H3K4me1, H3K4me3) known to be associated with active promoter and enhancer regions across 127 Roadmap tissues 11,27 , aggregated using the average (Avg) or maximum (Max) across tissues; 8 analogous Basenji annotations 16 , quantile-matched with DeepSEA annotations to lie between 0 and 1; and 9 BiClassCNN annotations that we trained to prioritize SNPs within non-tissue-specific features ( Table 1 and Methods). To assess whether the deep learning annotations provided unique information for disease, we conservatively included a broad set of coding, conserved, regulatory and LD-related annotations in our analyses: 86 annotations from the baseline-LD (v2.1) model 19 , which has been shown to effectively model LD-dependent architectures 28 ; 8 Roadmap annotations 11 (analogous to DeepSEA and Basenji annotations), imputed using ChromImpute 20 ; and 40 ChromHMM annotations 21,22 based on 20 ChromHMM states across 127 Roadmap tissues 11 (Table S1). In our analysis of allelic-effect (∆) annotations, we analyzed 16 non-tissue-specific deep learning annotations (8 DeepSEA∆ and 8 Basenji∆), again aggregated using the average (Avg) or maximum (Max) across tissues ( Table 1 and Methods).…”

“…Roadmap annotations 11 (analogous to DeepSEA and Basenji annotations) imputed using ChromImpute 20 , and 40 ChromHMM annotations 21,22 based on 20 ChromHMM states across 127 Roadmap tissues 11 , again aggregated using the average (Avg) or maximum (Max) across tissues (Table S1 and Methods). Of these, 4 Roadmap annotations and 17 ChromHMM annotations attained a Bonferroni-significant τ conditional on the baseline-LD model ( Figure S2, Figure S3 and Table S7).…”

“…Here, we evaluate the informativeness for complex disease of both variant-level and allelic-effect annotations at commonly varying SNPs constructed using two deep learning models previously trained on tissue-specific regulatory features (DeepSEA 13,15 and Basenji 16 ); we also train a new deep learning model to prioritize SNPs within non-tissue-specific features (BiClassCNN). We apply stratified LD score regression 5,19 (S-LDSC) to 41 independent diseases and complex traits (average N =320K) to evaluate each annotation's informativeness for disease heritability conditional on a broad set of coding, conserved, regulatory and LD-related annotations from the baseline-LD model 19 and other sources (imputed Roadmap and ChromHMM annotations 11,20,21,22 ); as a secondary metric, we also evaluate the accuracy of models that incorporate deep learning annotations in predicting disease-associated or fine-mapped SNPs 23,24 . We aggregate DeepSEA and Basenji annotations across all tissues in meta-analyses across all 41 traits, across blood cell types in meta-analyses across 11 blood-related traits, and across brain tissues in meta-analyses across 8 brain-related traits.…”