We are developing new computational methods to use genomic DNA methylation and chromatin modification data to understand the functional relationship between these marks and gene expression. We have a particular interest in applying new methods in machine learning and data mining to unravel potential functional consequences of epigenetic changes in human disease, including cancer.
ME-Class and ME-Class2 are computational approaches to integrate genomic DNA methylation and expression data to determine a list of genes for which DNA methylation and expression are associated. The principle behind the analysis is that we are interested in genes for which DNA methylation changes can be used to predict expression changes. These tools have modules to use feature importance and unsupervised clustering to determine which of the underlying methylation changes are important. Both ME-Class and ME-Class2 are available on GitHub.
Schlosberg CE, Wu DY, Gabel HW, Edwards JR. ME-Class2 reveals context dependent regulatory roles for 5-hydroxymethylcytosine. Nucleic Acids Res. 2019 Mar 18;47(5):e28. doi: 10.1093/nar/gkz001. PubMed PMID: 30649543; PubMed Central PMCID: PMC6412249.
Schlosberg CE, VanderKraats ND, Edwards JR. Modeling complex patterns of differential DNA methylation that associate with gene expression changes. Nucleic Acids Res. 2017 May 19;45(9):5100-5111. doi: 10.1093/nar/gkx078. PubMed PMID: 28168293; PubMed Central PMCID: PMC5435975.
WIMSi – Washington University Interpolated Methylation Signatures
WIMSi is a tool for discovering patterns of methylation that correlate with differential expression, with minimal prior assumptions about what patterns should exist in the data. We first represent differential methylation at each promoter as an interpolated curve, or methylation signature. This form allows us to compare methylation signatures using shape-similarity techniques. For regions with missing data, the interpolation also provides a reasonable guess for the methylation level. Using a shape-similarity metric known as the coupling distance (a discrete version of the Fréchet distance), we identify groups of genes with similar methylation signatures that also have corresponding expression changes. WIMSi is available from Sourceforge.
Vanderkraats ND, Hiken JF, Decker KF, Edwards JR. Discovering high-resolution patterns of differential DNA methylation that correlate with gene expression changes. Nucleic Acids Res. 2013 Aug;41(14):6816-27. doi: 10.1093/nar/gkt482. Epub 2013 Jun 7. PubMed PMID: 23748561; PubMed Central PMCID: PMC3737560.