Multi-group comparison doubts

[Alejandra Rodríguez]

Hello everyone,

I am currently working on a differential methylation analysis project and I have three groups of samples: normal, primary tumour, and metastasis samples. I aim to identify regions with significant methylation differences among these groups using the methylKit package.

I have:

Loaded my methylation data into a methylBase object.
Defined a treatment vector indicating the group each sample belongs to (0,1,2) and generated methylation object

myObject <- methRead(location=sample_files, header=FALSE, sample.id=sample_ids, assembly="hg19", treatment=treatment, pipeline="bismarkCoverage")

meth <- unite(myObject, destrand=FALSE, min.per.group=1L)

meth <- reorganize(meth, sample.ids=sample_ids, treatment=treatment)

Aggregated methylation data into 500bp tiles

meth_tiles <- tileMethylCounts(meth, win.size = 500, step.size = 100, cov.bases = 1)
Performed differential methylation analysis using calculateDiffMeth() on the tiled data.

diff_met <- calculateDiffMet(meth_tiles)

I understand that when there are more than two groups, calculateDiffMeth() calculates the methylation difference (meth.diff) as the difference between the maximum and minimum mean methylation values across the groups.

QUESTION 1:
How can I best interpret a meth.diff value in this context? For example, if meth.diff is 17.892, does it indicate a moderate difference in methylation between the most and least methylated groups?

QUESTION 2:

I did originally set it up so that I would perform pairwise comparisons to identify those significantly methylation different regions among my groups, but it was suggested to me that I performed a multi-group comparison where control would act as a baseline of sorts, and we would be able to see the signal from primary tumor and met more clearly.

As much as I have searched online, I cannot find a lot of information on how to do this. Is it common practice in methylation analysis?

The closest I have come to it has been performing this multi-group comparison, where I obtain a value that represents the difference between the group with the highest average methylation and the group with the lowest average methylation for a given region, but I cannot know which group is which.

Am I missing something?

[Alexander Blume]

Hi Alejandra,

Regarding your first question, as you already described, the "meth.diff" value alone doesn't specify which groups are involved or the direction of the change (i.e., which group has higher methylation). To interpret whether this is a moderate difference, you might consider the biological context and the typical range of methylation differences observed in similar studies. Generally, a difference of around 10-20% can be considered moderate, but this can vary depending on the system and context. Usually, we use 25% (together with a q-value of 0.01) as the cutoff to define DMCs/DMRs. 

Concerning the second question, I think the approach you described sounds reasonable, it is similar to what we did in a recent multip-group study where we used pairwise testing (https://academic.oup.com/nargab/article/5/2/lqad061/7209339#409595211). 

You can also start with the multi-group comparison to narrow down the regions of interest (described by Altuna: https://groups.google.com/g/methylkit_discussion/c/mTpy8I9dJUk/m/ZkUPDF7qAQAJ).

This would be a two-step process, where based on the multi-group comparison you define a subset of regions with minimal methylation difference that can be followed up with pairwise tests. 

Best,

Alex