Bioinformatics Seminar
Time: 11AM
Venue: Teams
5 July 2022
What is a DMR? The concept is in the calling
Tim PetersGarvan Institute
There is very little consensus between research groups on how differential methylation between cell types and/or disease states ought to be summarised and reported. Highly bespoke workflows dominate the research landscape, due in part to decentralised research groups each having subtly different notions of what constitutes a differentially methylation region (DMR), which is reflected in the bioinformatic tools they develop. We discuss the most desirable characteristics for DMR detection tools in a genome-wide context, and present a benchmarking study on both DML (differentially methylated loci) and DMR callers using highly realistic simulated whole genome bisulfite sequencing (WGBS) data that contains distribution parameters derived from a large-scale consortium dataset. We find that predictive performance depends on a combination of various methylation data parameters such as coverage and sample size, but find that WGBS counts transformed by limma-voom with library size adjustment is the winning strategy in the greatest number of cases. We use this result to backend and introduce an expanded version of DMRcate: an existing DMR detection tool for array data now able to call DMRs from WGBS. We compare DMRcate to a set of alternative DMR callers and find DMRcate and RADmeth are essentially equal best predictors of DMRs, but conclusively find DMRcate the fastest. We also find that DMR and DML predictive capacity increases as coverage increases, with diminishing gains but no appreciable plateau.