Motivation: Chromatin structure, including post-translational modifications of histones, regulates gene expression, alternative splicing and cell identity. ChIP-seq is an increasingly used assay to study chromatin function. However, tools for downstream bioinformatics analysis are limited and are only based on the evaluation of signal intensities. We reasoned that new methods taking into account other signal characteristics such as peak shape, location and frequencies might reveal new insights into chromatin function, particularly in situation where differences in read intensities are subtle.
Results: We introduced an analysis pipeline, based on linear predictive coding (LPC), which allows the capture and comparison of ChIP-seq histone profiles. First, we show that the modeled signal profiles distinguish differentially expressed genes with comparable accuracy to signal intensities. The method was robust against parameter variations and performed well up to a signal-to-noise ratio of 0.55. Additionally, we show that LPC profiles of activating and repressive histone marks cluster into distinct groups and can be used to predict their function.
Availability and implementation: http://www.cancerresearch.unsw.edu.au/crcweb.nsf/page/LPCHP A Matlab implementation along with usage instructions and an example input file are available from: http://www.cancerresearch.unsw.edu.au/crcweb.nsf/page/LPCHP.