Identifying Spatially Correlated Changes in Methylation Profiles Using Kernel Methods

Tom Mayo, Guido Sanguinetti, Gabrielle Schweikert

Research output: Contribution to conferencePosterpeer-review

Abstract

DNA methylation is an intensely studied epigenetic mark associated with many fundamental biological processes of direct clinical relevance. Bisulfite treatment of DNA followed by next generation sequencing provides quantitative methylation data at base pair resolution. However, statistical modelling of such data is challenging. Current approaches do not consider higher order features of the data, so that spatially correlated changes are ignored. A recent paper has shown that the shape of the methylation profile change is predictive of gene expression. Furthermore, parametric tests require high coverage and replication and are prone to overconfidence under high coverage conditions.
We introduce a non-parametric test, M3D, based on the maximum mean discrepancy to address such issues. M3D uses kernel methods to capture spatially correlated changes in methylation profiles and displays improved power over existing methods in challenging conditions. The method is freely available via Bioconductor as package M3D.
Original languageEnglish
Publication statusPublished - 2014
EventEighth International Workshop on Machine Learning in Systems Biology - Strasbourg, Austria
Duration: 6 Sep 20147 Sep 2014

Workshop

WorkshopEighth International Workshop on Machine Learning in Systems Biology
CountryAustria
CityStrasbourg
Period6/09/147/09/14

Fingerprint Dive into the research topics of 'Identifying Spatially Correlated Changes in Methylation Profiles Using Kernel Methods'. Together they form a unique fingerprint.

Cite this