DE analysis with [DESeq2] without replicates

Question

[john]

Hello,

This post is about using DESeq2 for DE analysis without replicates.

I have 10 individually collected dishes with cells. Time point 0h is not treated. Time point 2h is collected after 2h, time point 4h is collected after 4h etc.

So in total 10 time points c("h0","h2","h4","h6","h8","h10","h12","h14","h24","h48")

I don't have replicates- which is a problem. Could I simply take lets say the

h0 with h2(as replicate)

against

h4 with (h6, h8 as replicate)

And do a simple two groups comparison? So the idea is to use Michael Love and Simon Anders assumption that one can expect most genes as not differentially expressed. Like this I am trying to get some fake replicates.

What about I take 0h as one group and 2h-48h as second group? I mean this is obvious better than taking 0h and 2h as untreated and treated since then I don't have replicates at all.

I am obviously trying to make more out of my data than there really is but right now I can only work with what I have got.

Any thoughts on that?

[jing]

If the assumptions are not believable, the results will be no good.

One thing in my mind is to check literatures with similar experiment setup, and see what genes are most probably constitutively expressed at constant levels in your experiment. You can then use those for normalisation, or at least, for checking whether your assumptions are sound.

[john]

Yes jing,

thats probably a good idea! Are you currently working on a similar experiment by any chance?

[Biomonika (Noolean)]

And what would you compare to what if you had all the replicates? Are you interested in each step in time series or just final untreated/treated difference?

[john]

Hello Noolean,

In general I am interested in both the change over time and treated/untreated.

For the "over time change" I did the following A: [DESeq2] time series analysis. Linear and squared model please tell me if u know a better way!

However in this current post here, I am interested in treated/untreated.

Answer 1

[kangyueapril]

I don't think it is suitable to "make" it as replication by yourself. If you don't have replication you can't tell which results is reliable. But if you still want to use this data, maybe you can just use h0,h2 as a group and h10,h12 (or h12, h14) as a group. But just ignore rest of data.