How to perform differential gene expression analysis with only one replicate in each condition

Question

[Apex92]

Hi all,

I am trying to perform differential gene expression analysis with only one replicate in each condition (one Control vs one KD). Is there any reliable tool that I can use for this purpose? Or do you have any specific approach/suggestion in your mind that can be used without the need to use any tool?

I have already seen this post but I thought maybe new tools/approaches came out already so it is better to ask a similar question again.

Thanks a lot.

[Michael]

Just for clarification, this is a principle thing. So you could wait 1000 years and the answer will still be No.

Answer 1

[lieven.sterck]

Noop.

and moreover it does not have anything to do with tools or such. as it is written in the post you refer to it's statistics, without n > 1 there is nothing to reliably calculate the values you need to call something differentially expressed.

[Apex92]

Maybe it was not a good option to say "differential gene expression analysis" but rather `i want to find a reliable way where I can find some genes that seem to be different in the conditions. What about p-value free methods like this one ?.

[Michael]

I think it's more about the word "reliable" you used in the sense of having accuracy or even significance. No level of mathematical woo can change the fact that you cannot base a generalizable conclusion about a group or random effect on a single observation, and honestly, I would wish that tools wouldn't be advertised as if they could.

What you can do is to make a deliberately "unreliable" selection aka. screening, the simplest approach being to rank genes by the absolute difference in expression to get some candidates or use a z-score. If you chose the top-ranked genes and do some additional repeated measurements, e.g. qPCR, you could still discover something interesting. We have done something like that in a recent paper, where we didn't have replication of RNA-seq to start with, but then this requires an enormous amount of follow-up experiments to get through rendering the initial candidate selection almost a marginal part of the study. So, it is possible to such a screening method to pick some low-hanging fruit, but only when assuming it is exactly not reliable.