Experimental Design with DEseq2

Hi All,

I want to run a differential expression experiment with DEseq2, however I am a little confused regarding how to set up the formula. I have matched tumor/normal pairs and I want to contrast that to a third clinical variable (see table below).

Sample Tissue Clinical_Status

SampleA Tumor X

SampleA Normal X

SampleB Tumor Y

SampleB Normal Y

The biological question is: "Are there Differentially expressed genes in tumor samples according to clinical status?"

To answer this I could remove the normal samples all together in which case the design formula would be this: design = ~ Clinical_Status

However This does not seem optimal as I would have no idea if the genes would be DE in the normals as well and as such should be ignored. I could run the test again (for normals) and remove those DE genes from the original test however I believe there would be a multiple testing problem there.

I am thinking there is a way to test what I want all at once and design should be something like this: design = ~ Sample + Clinical_Status + Tissue:Clinical_Status

However I am not 100% sure, any help would be appreciated. apologies for the naivety of the question.

deseq2 experimental design r

• 2,530 views • link

updated 9.7 years ago by ivivek_ngs

• written 9.7 years ago by zlskidmore

1 answer

It should be a ~patient+ condition design ti effectively find out DE genes. For doing DE analysis you need to know genes that are up and down in Tumor w.r.t to normals. So why would you remove them. Unless you have batches and other variables that mask the effect you should have a fairly straight forward design.

When you say clinical status I assume it is either tumor or normal. Check for the links below. It seems you have 2 conditions with 2 replicates per condition. If they are paired then a paired test should be employed. Simply follow the links to get a better understanding or follow the manual which should also give a clear example of tumor/normal RNA-Seq paired data analysis. Unless you have other confounding factors this design should be pretty much straight forward.

link1

link2

• 0 views • link

written 9.7 years ago by ivivek_ngs

maybe i'm not understanding... what i'm looking for is slightly more complicated I think than ~ sample + clinical_status. I want to know only those genes differentially expressed based on clinical status in the tumor samples only (i.e. I want ~ sample + clinical status but i also want to remove anything that would have come out of that interaction if it was run on the normal samples only)

So there are 3 different things here, I have samples which have an associated tissue type (paired tumor/normals) and I have a clinical status, lets say response vs non response.

• 0 views • link

written 9.7 years ago by zlskidmore