What is a typical workflow to correlate methylation and expression data?

Question

[biohack92]

I've been able to find differentially methylated CpG positions using minfi along with differentially expressed genes using the limma package. What is the next step to determining the relationship between the expression (Agilent 4x44k) and methylation data (Illumina 450k)?

My thought process is as follows:

I have two groups, pre- and post- treatment. I would compare the mean delta beta values for the CpGs associated with each gene with the log fold changes from the expression analysis. If this thought process is correct, how do I make this comparison? If not, what would be the correct approach?

Appreciate any insight!

[Nicolas Rosewick]

I'm also interested in this kind of analysis. Maybe could you give us a feedback on your analysis. Thanks

Answer 1

[Shicheng Guo]

Another R package can be used is MethylMix: An R package for identifying DNA methylation driven genes.

source("https://bioconductor.org/biocLite.R")
biocLite("MethylMix")

or

source("http://bioconductor.org/biocLite.R")
biocLite("MethylMix")

[Nicolas Rosewick]

I do not see any part in the vignette talking about correlation between methylation and expression data. Maybe I missed something ?

Answer 2

[Charles Warden]

I have designed a pipeline integrate methylation and gene expression data, which was specifically designed around 450k data:

http://nar.oxfordjournals.org/content/41/11/e117.full

This package is also available in Bioconductor:

http://bioconductor.org/packages/devel/bioc/html/COHCAP.html

Are your expression and methylation data paired? If so, I think the best strategy is to look for a negative correlation between the intensity values and beta values. If not, then you can look for overlap in genes identified separately from the methylation and gene expression analysis.

[Charles Warden]

It looks like this is another option (again, if you have paired methylation and expression data):

http://bioinformatics.oxfordjournals.org/content/30/16/2360.full

[Vasu]

Hi Charles,

I have differentially expressed genes from Transcriptome analysis. And right now I'm working on Illumina methylation Data using EPIC (850k) data (Microarray analysis). I have the beta values.

So, now based on cutoff (beta value 0.2) I have some probes like cg18478105, cg09835024 etc... From the Annotation I also have their symbols.

Now I want compare the expression values from Transcriptome analysis with methylation levels of each gene. Can I use MethylMix or COHCAP for that? As my methylation data is from EPIC platform I didn't find any proper tool. When I looked into COHCAP it is only for 27k and 450k.

Can you please help me in this? Any advice is greatly appreciated. Thank you in Advance

[Shicheng Guo]

You can apply linear mixed model to make the inference between methylation and expression directly.

[Vasu]

linear mixed model ?

[Shicheng Guo]

Yes. You can try to fit Linear Mixed-Effects Models using lme4 package in R

[Bioinformatist Newbie]

Hi Charles,

Every time I try COHCAP.site to get differentially methylated sites between my control and diseased condition I get an empty file. I have tried it with beta values for BS data as well as for 5hmC data but in both cases results are same. That's what I am trying to do: filtered.sites <- COHCAP.site(sample.file, beta.table, project.name, project.folder, ref="Control", methyl.cutoff=0.7, unmethyl.cutoff = 0.3, paired=FALSE, delta.beta.cutoff = 0.05, pvalue.cutoff=0.05,fdr.cutoff=0.05, num.groups=2, plot.heatmap=TRUE, output.format = "txt")

[1] "Reading Sample Description File...."
[1] 178591     75
[1] 178591     74
[1] "Differential Methylation Stats for 2 Groups with Reference"
[1] 178591      5
[1] 178591     10

[1] 178591     10
[1]  0 10
[1]  0 10

I also tried to change cutoffs to 0.1 but still no DMS are detected. Can you guide me what I am doing wrong here. This is how my beta file looks like:

> beta.table[1:3,1:8]
      SiteID Chr       Loc    Gene                  Island      X00_10
1 cg00000029  16  53468112    RBL2 chr16:53468284-53469209  0.20818811
2 cg00000108   3  37459206 C3orf35                    <NA> -0.01900229
3 cg00000109   3 171916037  FNDC3B                    <NA> -0.01658023
      X00_13     X00_28
1 0.14817104 0.25321607
2 0.05034043 0.01979151
3 0.03844963 0.15055532

[Nicolas Rosewick]

You should post it as a new thread.

[Charles Warden]

Lowering the delta-beta cutoff will only help if the methylated thresholds are also changed. You could try setting both methyl.cutoff and unmethyl.cutoff equal to 0.3, or disable the use of methylated thresholds by setting unmethyl.cutoff=1 and methyl.cutoff=0.

Yes - in terms of posting etiquette, this should be new thread or you can post the issue on the COHCAP discussion group: https://sourceforge.net/p/cohcap/discussion