DEG analysis between 2 samples using edger

Question

[bioyas]

Hello everyone,

I have two samples under 2 different conditions. I would like to use edgeR to do DEG analysis between these 2 samples. I have read the edgeR manual and I know that if we have 2 or more biological replicates in both conditions we can use glmQLFit(). Or if we don't have biological replicates for our samples in experiment we can use exactTest() and manually set the dispersion.

Now I am dealing with the case that my sample1 has 2 biological replicates and sample2 is only one replicate. I want to compare these 2 samples to do DEG analysis. I went over the edgeR manual but I could not find such case being explained.

Do you have any idea how can I handle this?

Thanks

[Kevin Blighe]

I have tried, in my brain, just now, to support this type of comparison, which is a 2 versus 1 comparison, but I just cannot bring myself to do it. Unless this is simply a training exercise for you such that you can become familiar with coding in EdgeR, I just could not otherwise support this type of analysis in research; so, is it just for training purposes only?

Section 2.11 of the EdgeR manual more or less covers this. However, it is also stated:

Reliable estimation of dispersion generally requires replicates

Edit: keep in mind that I usually answer while thinking about things from a clinical perspective. As Gordon points out, it is not impossible to do the analysis that you want. I think that you would struggle to publish it though, depending on who is/are the reviewers.

[bioyas]

Thank you Kevin for your respond.

I understand that the result may not be reliable I do such analysis. I have seen section 2.11 of edger manual and I understand how to deal with the situation when the whole experiment is with replicates or if the whole experiment is without replicates. I know that reliable estimation of dispersion requires replicates. I just want to know if it is doable to have do this analysis when in our study we have samples with replicates and some samples without replicates. I don't know how to deal with the case that I have a mixture of these two types of samples. It can be a training purpose to see if it is feasible regardless of the reliability of results. But I don't know how to do it. Can you please guide me on this.

Thanks

[Gordon Smyth]

This question has been answered a number of times, for example:

Differential expression: replicates in one condition, no replicates in the other

Answer 1

[Gordon Smyth]

Now I am dealing with the case that my sample1 has 2 biological replicates and sample2 is only one replicate.

You have at least some biological replication so you can use edgeR. You do not need replication in every group.

I want to compare these 2 samples to do DEG analysis. I went over the edgeR manual but I could not find such case being explained.

No, because there's nothing to explain. It works in edgeR exactly the same as any other two-group problem.

You can use any of the edgeR pipelines glmQLFit() or glmLRT() or exactTest(), but we recommend glmQLFit().

Obviously the n=2 vs n=1 case is minimal and more replicates would be highly desirable. But edgeR will do the best that can be done and will yield results that are at least defensible. Use edgeR QL for the n=2 vs n=1 case is specifically discussed in the comment section of the following paper. I show in my reply to Conrad Burden that edgeR gives reasonable error rate control even in this minimal case:

https://f1000research.com/articles/5-1438

[Kevin Blighe]

Thanks Gordon