A Deep Learning Model for Epigenomic Studies

Abstract

Epigenetics is the study of heritable changesin gene expression that does not involve changes to theunderlying DNA sequence, i.e. a change in phenotype notinvolved by a change in genotype. At least three mainfactor seems responsible for epigenetic change including DNAmethylation, histone modification and non-coding RNA, eachone sharing having the same property to affect the dynamicof the chromatin structure by acting on Nucleosomes position. A nucleosome is a DNA-histone complex, where around150 base pairs of double-stranded DNA is wrapped. Therole of nucleosomes is to pack the DNA into the nucleusof the Eukaryote cells, to form the Chromatin. Nucleosomepositioning plays an important role in gene regulation andseveral studies shows that distinct DNA sequence featureshave been identified to be associated with nucleosomepresence. Starting from this suggestion, the identificationof nucleosomes on a genomic scale has been successfullyperformed by DNA sequence features representation andclassical supervised classification methods such as SupportVector Machines, Logistic regression and so on. Taking inconsideration the successful application of the deep neuralnetworks on several challenging classification problems, inthis paper we want to study how deep learning network canhelp in the identification of nucleosomes