MIG Seminar Series - Jovana Maksimovic - Gene set testing for differentially methylated regions
Gene set testing for differentially methylated regions
DNA methylation is the most extensively studied epigenetic modification. It is necessary for normal embryonic development and is often altered in disease, thus providing valuable insight into many biological mechanisms. Illumina’s HumanMethylation BeadChips remain the most popular platform for assaying methylation in humans. Their newest “EPIC” BeadChip interrogates the methylation status of >850,000 CpGs across the human genome. CpGs are richly annotated and are heavily biased towards genes and other regions of interest such as enhancers.
Typically, a probe-level analysis looks for associations between methylation levels at individual CpGs and a factor of interest e.g. disease status. We have previously shown that genes with more CpG probes on the Illumina array are more likely to have significantly differentially methylated CpGs, biasing downstream gene set analysis, which is often used to interpret long lists of significant CpGs. We developed the GOmeth method to account for this bias, for probe-level analyses. However, region-level analyses, which identify correlated methylation patterns between several spatially adjacent CpGs, can yield more functionally relevant results.
Here we demonstrate that the same bias affects gene set testing of region-level analyses and have extended GOmeth accordingly. We show that GOregion identifies more biologically meaningful gene sets than a naive hypergeometric test. We also demonstrate that a GOregion analysis can distil more focused gene sets than a probe-level gene set analysis of the same data, which is highly dependent on the significance cut offs used. GOregion is compatible with the results of any software for finding differentially methylated regions, which can be expressed as a ranged data object. It is also highly efficient and can test a variety of gene sets such as GO categories, KEGG pathways or any list of custom gene sets. GOregion is available as a function in the missMethyl Bioconductor R package.
Dr Jovana Maksimovic, Postdoctoral Researcher, Bioinformatics Group, MCRI
Dr Jovana Maksimovic
Postdoctoral Researcher, Bioinformatics Group, MCRI
Murdoch Children's Research Institute
After completing a Bachelor of Science (Honours)/Bachelor of Bioinformatics at La Trobe University, Jovana was a graduate with the Department of Primary Industries (DPI) for 2 years. During this time, she was involved in many diverse research projects and developed an interest in the biology of lactation. She started her PhD at Monash in 2007 on a DPI project investigating the expression and regulation of a gene family involved in the production of a subset of milk oligosaccharides that are of particular interest in infant nutrition. Jovana currently works as a postdoctoral researcher in the bioinformatics group at MCRI, working with A/Prof Alicia Oshlack, where she is focused on the analysis of gene expression and epigenetic data and the development of new computational and statistical methods. Jovana has published multiple analysis methods for methylation array data and jointly maintains an R Bioconductor package for methylation array analysis: missMethyl. She was one of the organisers of the inaugural Australian Bioinformatics Conference (ABiC) in 2014 which paved the way for the annual Australian Bioinformatics and Computational Biology Society (ABACBS) conference. Jovana is also the current postdoctoral representative on the ABACBS executive committee.