Hi all,
Pretty straight forward question, I've not seen this approach so ...red flag.
My data set:
The primary question is to find differentially expressed genes (DEGs) within each tissue, but I would also like to see if there is an overall set of DEGs across tissues.
Question: Can I treat "tissue" as a covariate? That is, try to get DESeq2/EdgeR to treat it as a batch effect and reduce the effect of the difference between tissues so that I can get an overall picture of the difference between the 2 conditions regardless of tissue type. It seems intuitively more powerful than getting DEGs for each tissue and visually comparing lists
Thanks, and apologies if this is a complete no-no. Kenneth.
I don't think this approach will give you what you want. Combining multiple tissues in the way you are suggesting will result in a positive set consisting of genes strongly differentially expressed in one single tissue, genes moderately differentially expressed in a few tissues, and genes mildly differentially expressed across all tissues.
It seems to me the null that you want is "0 change in one or more tissues" (so that the alternative is a change in all tissues) -- and the easiest way to get that is (indeed) to generate the venn diagram (or UpSet) of DE genes.
And that's before getting into the issue that including tissue as a covariate only adjusts for changes in mean expression -- but the variance will also differ (substantially) across tissues, which leads to a violation homoskedasticity assumptions.
Thanks LChart, I accept that, I'll wait to see if there's any more who want to input and then accept the answer
I would not put totally different tissues in the same DESeq object. I don't think that's going to do good things for normalization or dispersion estimates.
Make each tissue its own object, make your design batch + condition and compare conditions to each other. Then compare gene lists between tissues.
great guidance ... thank you
Log in to answer this question.
