?Design of DE analysis when one sample has no repeat

Question

[drodavis0]

Hi! Briefly looking for some advice on what I can do with the following problem (I've tried searching this and other forums for advice, but can't seem to find any direct answers).

I'm performing the bioinformatic analysis for another group's chip-seq data, which contains 2 replicates for a control and 1 sample for a treatment group. Due to various factors, I'm not able to obtain a repeat of the treatment group, which would otherwise be the solution to my problem.

I'm trying to think of the best way of analysing this with differential enrichment tools (e.g. with something like edgeR). Does anyone have any advice on the best way I could approach this? I wondered if one approach might be to treat each sample as as if they were all unique and thus there were no replicates, but then is this wasting potentially useful information in the repeat control sample (in terms of how the tool I use would treat the data)?! Do tools like DESEQ2 or edgeR still work with repeats in the control group, but only 1 sample in the treatment group?

I really appreciate any advice or tips people can offer, Thanks!

[drodavis0]

Thanks very much for your advice!

Answer 1

[i.sudbery]

I'm not sure about DESeq2 but edgeR can, in theory, handle this. I think. Or at least it used to, I don't know if it does anymore. The idea is to assume that the dispersion is the same in control as it is in treatment, and to lean heavily on borrowing information between genes of similar expression level. I think this sort of analysis is probably quite likely to lead to a higher than desired false positive rate for peaks that are innately highly variable, but it might suffice for hypothesis generation purposes.

[Antonio R. Franco]

One of my students ran exactly this approach using edgeR and it worked beautifully. The GO enrichment information we got was very much in agreement to what we expected. In the absence of replicates, we believe this is the way to go