Any suggestion of tool for differential expression test when one replicate for one condition and no replicate for the other?

Question

[thejustpark]

I am trying to get differentially expressed genes for this experiment

where one condition has one biological replicate (two samples) and the other condition doesn't have one (one sample).

I read many discussions about how bad it is to do differential expression test for no replicate samples.

But I am still unclear how to do the test for my situation (I believe my situation is better than completely no replicate case).

Can you guys suggest me a good tool for this analysis?

Thanks,
HJ.

[Devon Ryan]

Hello thejustpark!

It appears that your post has been cross-posted to another site: SeqAnswers.

This is typically not recommended as it runs the risk of annoying people in both communities.

[thejustpark]

Just one more comment: the biologists said that the two conditions are expected to be not much different, only affecting a number of genes. That is another reason I think it would be OK not to require additional replicate at least at this time point.

[thejustpark]

Devon,

I like your solution. I will give it a try.

Thank you for the suggestion.

And about cross-posting, I should have noticed that before.

My apologies. I won't do that again.

HJ.

Answer 1

[Devon Ryan]

Your situation is a bit better than no replicates at all. Any of the normal tools (DESeq2, edgeR, limma, cuffdiff) should work. Keep in mind that you have really low power to be able to detect anything, so don't blindly accept any DE findings without replicating them with another technology.

BTW, regarding your comment, if they're only expecting a small difference between the groups then they're likely to be disappointed in the analysis. The smaller the difference, the more samples you need to find it. This is true for all studies that I can think of (i.e., this isn't a bioinformatics-specific thing).

[Irsan]

I would like to add that chapter 2.10 in the edgeR user guide advices about what to when you don't have replicates. For me, a useful solution has been to define a set of housekeeping genes (genes for which you assume they have comparable expression levels between all your samples) and estimate the common dispersion based on those genes only. Then proceed as explained in the edgeR manual