Batch effect in Methylkit and SVA

[jussalmi]

Hello!

I have 6 RRBS samples that I am analyzing in MethylKit. After looking at the clustering and PCA plots it seems that there are batch effects. The 6 samples were done at 3 different times, each time one control and one case. The samples cluster according to these batches.

How can I remove the batch effects with SVA or other tools? I looked at the assocComp function and there were 3 PCA components that seemed to be correlated with batches at the 0.1 - 0.2 level. I don't trust that removing those 3 components removes just the batch effects.

SVA is recommended in the documentation, but I'm not an expert in R and I'm not sure how to use with MethylKit.

[jnhutchinson]

Did you ever figure this out? I'm in the same boat with batch effects.

I've seen other posts claiming people use SVA on logit transformed values, but I'm not sure how I would - or if I should - incorporate the SVA transformed data back into the methylKit workflow.

[jussalmi]

Radmeth

No I didn't. I moved my analysis to RADmeth/Methpipe. In RADmeth you can define covariates in the design matrix to take care of batch effects.

[jnhutchinson]

Thanks! That's really helpful. I'll check it out.

[jussalmi]

Example

Here's a clarifying text I got from the author of RADmeth (I had 15 samples in three batches):

"First of all, you are absolutely correct. To account for batch effects you can add a few more columns to the design matrix. I have attached an example of such a matrix. It describes a dataset containing 15 samples. The samples are grouped into 3 batches. This is encoded by columns base, batch1, and batch2. (Note that you only need two extra columns to specify three groups: batch1 and batch2 permit methylation levels of two groups of samples to deviate from the third, base group; this gives you three total groups that correspond to batches; see what I mean?). The last column, testfact, defines a factor level that you would use for testing.


base batch1 batch2 testfact
sample01 1 0 0 0
sample02 1 0 0 0
sample03 1 0 0 0
sample04 1 0 0 1
sample05 1 0 0 1
sample06 1 1 0 0
sample07 1 1 0 0
sample08 1 1 0 0
sample09 1 1 0 1
sample10 1 1 0 1
sample11 1 0 1 0
sample12 1 0 1 0
sample13 1 0 1 0
sample14 1 0 1 1
sample15 1 0 1 1
"