Question: another contrast question - edgeR + scRNA-seq + timecourse
gravatar for regan.hayward
18 months ago by
regan.hayward0 wrote:

I've been tasked with looking at DE genes in a scRNA-seq dataset. It's a timecourse experiment with three timepoints (3, 6 & 12 hrs) with a control and treatment at each time. There was an issue with the 3hr control cells and they are not being used. I clustered the cells and identified that all of the 6 and 12 hr cells cluster together and the 3hr cells cluster together separately.

What I'm trying to look at now is the difference between the two clusters taking into consideration the control cells.

So, a design like this:

3hr treatment - ((12 treatment - 12 control) + (6 treatment - 6 control))


Here is the code:

design <- model.matrix(~0+ group)
# groups have been set up like this: group_12T, group_12C, group_6T etc
dge <- estimateDisp(dge, design = design)
fit <- glmFit(dge, design)
con <- makeContrasts(group_3T-((group_12T - group_12C)+(group_6T - group_6C)), levels=design)
   Levels      group_3T - ((group_12T - group_12C) + (group_6T - group_6C))
group_12T                                                                -1
group_12C                                                                 1
 group_3T                                                                 1
 group_6T                                                                -1
 group_6C                                                                 1
fit = glmLRT(fit, contrast = con)


When I set this up and run it, the results are somewhat unusual biologically speaking.

My question is, have I made any mistakes with my process above? or is there a more suitable process of events I should be following?

Thanks in advance!

ADD COMMENTlink modified 18 months ago by Gordon Smyth39k • written 18 months ago by regan.hayward0
Answer: another contrast question - edgeR + scRNA-seq + timecourse
gravatar for Gordon Smyth
18 months ago by
Gordon Smyth39k
Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia
Gordon Smyth39k wrote:

No, the contrast you have taken doesn't really make sense. You're not testing for differential expression in the usual sense.

Generally speaking, the numbers that make up the contrasts should add to zero. You contrast doesn't add to zero because you have three 1's but only two -1's. You are in effect assuming that the 3 hour controls have zero expression, just because you didn't measure them, but that is not a valid assumption.

It makes sense to form any of the following contrasts:

(group_12T - group_12C) + (group_6T - group_6C)
group_3T - group_6T
group 3T - group_12C
group 3T - 0.5 * (group_6T + group_12T)

but not the one you have.

ADD COMMENTlink modified 18 months ago • written 18 months ago by Gordon Smyth39k

I see - I didn't realise I was making that assumption. Thanks for the explanation and quick response!

ADD REPLYlink written 18 months ago by regan.hayward0
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 452 users visited in the last hour