Multisample DMR calling

Question

[Aerval]

Hi everyone,

does anyone of you have recommendations for inter-multisample DMR (differentially methylated region) calling. I have WGBS methylation data for multiple cell populations and want to identify DMR specific for any group of them. For example DMRs specific for Lung and Heart in difference to Skin + DMR specific for Lung in difference to Heart and Skin + DMR specific to Heart in difference to Skin and Lung. For more populations you obviously get more comparisons. Now are most DMR callers only for pairwise comparisons. DSS-general and RadMeth offer group wise comparisons which seems nice but would need all pair-wise groupings and especially always includes all samples. Are there any alternatives?

Alternatively, we are currently doing all pairwise comparisons, then try and merge overlapping DMR and later reassociate the DMR generated this way to different cell types (currently via kmeans but looking into other things). Unfortunatly, I have to assume to lose information that way when a DMR is part specific to one cell type and part specific to multiple cell types. If anyone has suggestions or solved similar problems I'd love to hear some input?

Thank you,

[WouterDeCoster]

Please avoid the use of abbreviations, definitely without explaining them.

Answer 1

[Devon Ryan]

Give metilene a go, I've had some nice experiences with it.

Nothing does exactly what you want for a few reasons. For grouped comparisons (e.g., DMRs between skin and both heart and lung) you can just put heart and lung together in a group. This presumes you don't consider cases where heart, lung, and skin have 100%, 0% and 50% methylation, respectively, to be DMRs, of course.

[Aerval]

Thanks, I did not no metilene and will look into it.

Yeah, I presumed so, thats why I asked whether anyone has developed a solution for it. Thanks anyways.

[331885030]

Hi Devon I have downloaded the software metilene, but I have not found the data type of input file, could you please send me an example for the details of input data format ? my email [e-mail], thank you in advance.

Answer 2

[hongbo919]

Hi Aerval, we have developed a new software (SMART-BS-Seq) for de novo identification of DMRs from various BS-Seq data including WGBS. I think this software is what you are looking for. SMART-BS-Seq can identify the group-specific DMRs across multiple groups without pairwise comparisons. Detailed information can be found at http://fame.edbc.org/smart/ Best wishes! :)

[Aerval]

Thanks, that looks interesting. I am not working on that particular project anymore but will forward your link to the people still working on it.