Inferring chromatin states with stochastic autoencoder

Mechanisms of DNA compactization are efficient and complex, however despite their immense influence upon transcription the definite relationships between a plethora of epigenetic factors haven’t been fully disentangled. Histones undergo various post-translational modifications which alter nucleosome-DNA and nucleosome-nucleosome interactions, and hence the chromatin state. At least two chromatin states are easily identifiable, these are euchromatin, loose, transcriptionally active part of chromatin accessible by enzymes and transcriptional factors and heterochromatin which has compact and dense structure and is transcriptionally inactive. Epigenomic data for the majority of cell lines are very limited. Hence we attempt to construct a simple, universal model which would derive chromatin states based upon six histone modifications which should be present by specification in every reference epigenome of the project ENCODE.