Question: DESeq2: paired samples + controls
0
gravatar for guidobarzaghi
5 weeks ago by
University of Copenhagen
guidobarzaghi20 wrote:

Hi fellow DESeq'ers!

I am trying to figure out how to model a DEA with n paired samples (so n conditions + n controls, where the two sample sets come from the same individuals) plus two unrelated controls. So overall I am dealing with n condition + n+2 controls.

From the DESeq2 Vignette I could read that a paired experiment should be modeled as ~subject+condition, but I am wondering if this could work in my case as well.

Many many thanks in advance!

deseq2 • 77 views
ADD COMMENTlink modified 5 weeks ago by Michael Love22k • written 5 weeks ago by guidobarzaghi20
Answer: DESeq2: paired samples + controls
0
gravatar for Michael Love
5 weeks ago by
Michael Love22k
United States
Michael Love22k wrote:

In DESeq2 which uses fixed effects, the unpaired samples do not contribute to more information about the condition effect, because they do not have a corresponding treatment sample. So you can leave them out.

There is another method in limma-voom, called duplicateCorrelation(), which allows you to include them, as it does some extra modeling on the degree of correlation between the paired samples.

ADD COMMENTlink written 5 weeks ago by Michael Love22k

Wow, that was quick (thanks a lot for that).

I am only unsure about the meaning of "fixed effects", could you briefly elaborate on that?

ADD REPLYlink written 5 weeks ago by guidobarzaghi20

Maybe that's not very useful for me to have added, but I mean it to contrast against random or mixed effects modeling which can be more flexible for dealing with correlations among units.

https://en.wikipedia.org/wiki/Mixed_model

ADD REPLYlink written 5 weeks ago by Michael Love22k

Perfect, I think I got my head around this concept a bit more.

My final doubt then is whether adding those two external controls would somehow mislead the analysis or would it just add no information.

ADD REPLYlink written 5 weeks ago by guidobarzaghi20

In this design it just adds no information. I don’t think it produces any problem for the inference. I know this is the case for linear modeling at least.

ADD REPLYlink written 5 weeks ago by Michael Love22k

Alright, thanks a lot for you help!

ADD REPLYlink written 5 weeks ago by guidobarzaghi20
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 289 users visited in the last hour