Differentially methylated regions (DMRs) for a multi group comparison

Question

[ResearchR]

Dear all,

I am dealing with Whole-Genome-Bisulfite data and currently I am interested in DMRs. I have six tumor types and each tumor type has more than two subgroups. Since I would like to perform a intra- and inter-tumor analysis, a pairwise comparison for diff. methylated regions between tumors and groups within a tumor type is combinatorical not feasible. I search the web for solutions and found this paper, but the code, unfortunately, is not publicly available yet. Does anyone of you guys know a software package, which can detect DMRs in a mutligroup setting?

A example could look like: GBM vs. Medulloblastoma vs. ATRT vs. PA (for tumors)

Thanks for your help and best wishes,

Chris

[Kevin Blighe]

It looks like they built a multinomial Bayesian logistic regression model (you can use bayesglm in R) but did lots of other stuff on top of this. It's difficult to follow precisely the methods.

You could just try a lasso-penalised multinomial regression model, if you wanted: A: How to exclude some of breast cancer subtypes just by looking at gene expressio

You can also always just resort to the 'old reliables' of pairwise Wilcoxon Signed Rank tests (pairwise comparisons between your tumour types), or a non-parametric ANOVA (Kruskal-Wallis)