Learning chromatin states with factorized information criteria

Abstract

In this study, we propose a method to estimate the chromatin states indicated by genome-wide chromatin marks identified by NGS technologies. The proposed method automatically estimates the number of chromatin states and characterize each state on the basis of a hidden Markov model (HMM) in combination with a recently proposed model selection technique, factorized information criteria. The method is expected to provide an unbiased model because it relies on only two adjustable parameters and avoids heuristic procedures as much as possible. Computational experiments with simulated datasets show that our method automatically learns an appropriate model, even in cases where methods that rely on Bayesian information criteria fail to learn the model structures. In addition, we comprehensively compare our method to ChromHMM on three real datasets and show that our method estimates more chromatin states than ChromHMM for those datasets.

Systematic integration of GATA transcription factors and epigenomes via IDEAS paints the regulatory landscape of hematopoietic cells

“…4,38 A current approach for describing epigenetic landscapes is genome segmentation, 39 states to genomic segments exhibiting unique patterns of chromatin marks. Existing genome segmentation tools [39][40][41][42] were developed primarily for segmenting the epigenomes of single cell types. Although genomes from different cell types may be concatenated together, such an approach ignores the position-specific epigenetic events conserved across related cell types.…”

Systematic integration of GATA transcription factors and epigenomes via IDEAS paints the regulatory landscape of mouse hematopoietic cells

“…A current approach for describing epigenetic landscapes is genome segmentation (39,40), which assigns states to genomic segments exhibiting unique patterns of chromatin marks. Existing genome segmentation tools (39)(40)(41)(42) were developed primarily for segmenting the epigenomes of single cell types. Although genomes from different cell types may be concatenated together, such an approach ignores the position-specific epigenetic events conserved across related cell types.…”

Characterization of the human thyroid epigenome

“…Two popular model selection methods are the Bayesian information criterion (BIC) and the Akaike information criterion (AIC), but a weakness of these selection methods is that they tend to favor higher number of states that are biologically more difficult to distinctly interpret and does not capture sufficiently distinct interactions. Another metric for selecting the number of states produced by ChromHMM is the Factorized Information Criterion (FIC) proposed by Hamada et al (2015). However, FIC-HMM indicated more estimated chromatin states than what was selected for by the original ChromHMM analysis done by Ernst et al (2011) and thus are again biologically more difficult to distinctly interpret.…”