Gaussian derivative wavelets identify dynamic changes in histone modification

Epigenetic landscapes reveal how cells regulate genes in a cell-type or condition specific manner. Genome-wide surveys using histone modification showed cell-type specific regulatory regions. A number of computational methods were designed to identify cell-type specific regulatory regions using epigenome data. Most of them were designed to identify the enrichment of histone modification or their changes. However, they did not consider the shape of epigenetic signals, which represents the condition for protein binding at gene regulatory regions. We present a computational method to detect epigenetic changes using the shape of the signals for histone modification. Employing a Gaussian Derivative Wavelet (CGDWavelet) approach, the proposed method models a nucleosome with a Gaussian and detects the peak and the edges of the Gaussian. Using the detected parameters across two samples, CGDWavelet classifies epigenetic changes. We applied CGDWavelet to the histone modification data from mouse embryonic stem cells (mESCs) and neural progenitor cells (mNPCs) and identified four groups of epigenetic changes. Associating each group with gene expression, we found that gene expression is affected by chromatin structure as well as the intensity of histone modification. We found that Smad1, Sox2 and Nanog but not Oct4 bind to the epigenetically variable regions for H3K4me3. Software is available at http://wonk.med.upenn.edu/CGDWavelet