Cuffdiff time series with two groups

[sindrle]

Hi!
I have 4 sample groups. Diabetics and healthy at two time point, before and after intervention.

So, how can I analyse changes in gene expression in the diabetic groups, CONTROLLED for by the healthy group.

This means I have two discrete groups, but at two timepoint.

How to tell this to Cuffdiff?

Thanks!

[dpryan]

You can't, you'll have to use something else. Cuffdiff is only useful for very basic designs.

Edit: I should note that you could make the pair-wise comparisons with cuffdiff, but that's not what you want. Give DESeq2/edgeR/limma a try, they're MUCH more flexible.

[sindrle]

Wow. Thats a bummer.

What do you suggest? Im currently testing EdgeR.

[dpryan]

I'm quite happy with DESeq2. I've used edgeR again recently but have been a bit unhappy with it calling things significant due simply to an outlier sample (DESeq2 flags these on a per-gene basis, though I've also been bitten by this once in a partial knock-out dataset).

[sindrle]

Ok, sounds like DESeq is the way to go.

Thanks a lot!

Im thinking of running Cuffdiff 2.1.1, EdgeR & DEseq. Then compare everything and decide what to use in the publication.

[dpryan]

Im thinking of running Cuffdiff 2.1.1, EdgeR & DEseq. Then compare everything and decide what to use in the publication.

I suspect that's a pretty common route people take. Just make sure to validate a few candidates and then base your tool of choice on that rather than simply which gives the bigger list of DE genes!

[sindrle]

Thanks for the advice!
Any suggestion on candidates btw? I already have 5 genes tested with qPCR.

[dpryan]

Try to pick a couple that aren't called DE by all of them. That should help determine which of the models better fit your dataset.

[sindrle]

Thats a great tip!!
Thank you.

[sindrle]

Can you please guide me on using DEseq2 for this purpose?
I dont quite understand how to input:

Healthy at baseline, healthy at timepoint 2
&
Normals at baseline, normals at timepoint 2

[dpryan]

Can you please guide me on using DEseq2 for this purpose?
I dont quite understand how to input:

Healthy at baseline, healthy at timepoint 2
&
Normals at baseline, normals at timepoint 2

There are two ways one could look at that, I'll just give you one. Suppose you had 8 samples evenly divided into timepoints and normal/healthy groups:

status <- factor(c(rep("healthy",4), rep("normal",4)), levels=("normal", "healthy")
timepoints <- factor(c(rep(c(1,2), 4)))
des <- formula(~timepoints+status)

You can then use "des" as the design for your experiment. Alternatively, swap a "*" for the "+" on the last line to include an interaction term, which you probably want.

[sindrle]

You are awsome!
Some day I hope Ill repay in some way