edgeR batch correction

0

Entering edit mode

3.8 years ago

barix ▴ 20

Hi all,

I have 2 groups (experiment vs control) 4 samples per each group. The setup is like this:

enter image description here

I did MDS plot to see how my samples are clustered. enter image description here

Considering that experimental (hypoxic) replicates are far from each other, do I need to do batch correction here? If yes, what would be best approach here? I would appreciate any help/explanation of what's happening here. Thank you!

edger batch • 1.0k views

ADD COMMENT • link 3.8 years ago by barix ▴ 20

0

Entering edit mode

Do you have batches? With so few samples, I think you just have to roll with what you've got. You have variance in your treatment, and you can't magic that away.

ADD REPLY • link 3.8 years ago by swbarnes2 14k

1

Entering edit mode

You see grouping of 2/3 away from 1/4 regardless of treatment, so this could be a batch effect. @OP, try to remove the batch effect from the logcpms using limma::removeBatchEffect as a diagnostic and then repeat the MDS plot on these corrected data. If this shows better clustering then you may consider adding the batch variable (here that would be something like paste0("batch", c(1,2,2,1,1,2,2,1)) into the design.