MeDIP-HMM

by Michael Seifert, Sandra Cortijo, Francois Roudier, and Vincent Colot

Description

Motivation

Methylation of cytosines in DNA is an important epigenetic mechanism involved in transcriptional regulation and preservation of genome integrity in a wide range of eukaryotes. Immunoprecipitation of methylated DNA followed by hybridization to genomic tiling arrays (MeDIP-chip) is a cost-effective and sensitive method for methylome analyses. However, existing bioinformatic methods only enable a binary classification into unmethylated and methylated genomic regions, which limits biological interpretations. Indeed, DNA methylation levels can vary substantially within a given DNA fragment depending on the number of contained methylated cytosines. Therefore, a method for the identification of more than two methylation states from MeDIP-chip data is highly desirable.

Results

Here, we present a three-state Hidden Markov Model (MeDIP-HMM) for analyzing MeDIP-chip data. MeDIP-HMM utilizes a higher-order state-transition process improving modeling of spatial dependencies between chromosomal regions, allows a simultaneous analysis of replicates, and enables a differentiation between unmethylated, methylated and highly methylated genomic regions. We train MeDIP-HMM using a Bayesian Baum-Welch algorithm integrating prior knowledge on methylation levels. We apply MeDIP-HMM to the analysis of the Arabidopsis root methylome and systematically investigate the benefit of using higher-order HMMs. Moreover, we also perform an in-depth comparison study to existing methods and demonstrate the value of using MeDIP-HMM by comparisons to current knowledge on the Arabidopsis methylome. We find that MeDIP-HMM is a fast and precise method for the analysis of DNA methylation data enabling the identification of distinct DNA methylation levels. These results suggest that MeDIP-HMM could also be useful for analyses of other methylomes.

Paper

The paper MeDIP-HMM: Genome-wide identification of distinct DNA methylation states from high-density tiling arrays has been submitted to Bioinformatics.

Download

• The Arabidopsis root methylome data set and a JAR file containing the MeDIP-HMM will soon be available.