Question

edgeR time series DMSO correction

0

Entering edit mode

aec ▴ 90

@aec-9409

Last seen 3.9 years ago

Dear all,

I have a time series RNA-seq experiment with a control at time 0 and would like to test the inhibitors effect over time (4h, 24h) but also want to correct for DMSO effect for each time (4h, 24h).

This is what I have:

sample	replica	cell	time	condition	cell_time_condition
1	R1	cell_line1	0h	BaseLine	cell_line1.0h.BaseLine
2	R1	cell_line1	4h	DMSO	cell_line1.4h.DMSO
3	R1	cell_line1	4h	INHIBITOR1	cell_line1.4h.INHIBITOR1
4	R1	cell_line1	4h	INHIBITOR2	cell_line1.4h.INHIBITOR2
5	R1	cell_line1	4h	COMBO	cell_line1.4h.COMBO
6	R1	cell_line1	24h	DMSO	cell_line1.24h.DMSO
7	R1	cell_line1	24h	INHIBITOR1	cell_line1.24h.INHIBITOR1
8	R1	cell_line1	24h	INHIBITOR2	cell_line1.24h.INHIBITOR2
9	R1	cell_line1	24h	COMBO	cell_line1.24h.COMBO
10	R2	cell_line1	0h	BaseLine	cell_line1.0h.BaseLine
11	R2	cell_line1	4h	DMSO	cell_line1.4h.DMSO
12	R2	cell_line1	4h	INHIBITOR1	cell_line1.4h.INHIBITOR1
13	R2	cell_line1	4h	INHIBITOR2	cell_line1.4h.INHIBITOR2
14	R2	cell_line1	4h	COMBO	cell_line1.4h.COMBO
15	R2	cell_line1	24h	DMSO	cell_line1.24h.DMSO
16	R2	cell_line1	24h	INHIBITOR1	cell_line1.24h.INHIBITOR1
17	R2	cell_line1	24h	INHIBITOR2	cell_line1.24h.INHIBITOR2
18	R2	cell_line1	24h	COMBO	cell_line1.24h.COMBO
19	R3	cell_line1	0h	BaseLine	cell_line1.0h.BaseLine
20	R3	cell_line1	4h	DMSO	cell_line1.4h.DMSO
21	R3	cell_line1	4h	INHIBITOR1	cell_line1.4h.INHIBITOR1
22	R3	cell_line1	4h	INHIBITOR2	cell_line1.4h.INHIBITOR2
23	R3	cell_line1	4h	COMBO	cell_line1.4h.COMBO
24	R3	cell_line1	24h	DMSO	cell_line1.24h.DMSO
25	R3	cell_line1	24h	INHIBITOR1	cell_line1.24h.INHIBITOR1
26	R3	cell_line1	24h	INHIBITOR2	cell_line1.24h.INHIBITOR2
27	R3	cell_line1	24h	COMBO	cell_line1.24h.COMBO

The comparisons I am interested would be:

0h vs 4h inhibitor1

0h vs 4h inhibitor2

0h vs 4h COMBO

...

but correcting for the DMSO effect.

Here a piece of the code:

design <- model.matrix(~0+group, data=y$samples)

my.contrasts <- makeContrasts(

cell_line1_0h_BaseLine_vs_cell_line1_4h_inhibitor1=(cell_line1.4h.inhibitor1-cell_line1.0h.BaseLine)-(cell_line1.4h.DMSO-cell_line1.0h.BaseLine),
cell_line1_0h_BaseLine_vs_cell_line1_4h_inhibitor2=(cell_line1.4h.inhibitor2-cell_line1.0h.BaseLine)-(cell_line1.4h.DMSO-cell_line1.0h.BaseLine),
cell_line1_0h_BaseLine_vs_cell_line1_4h_COMBO=(cell_line1.4h.COMBO-cell_line1.0h.BaseLine)-(cell_line1.4h.DMSO-cell_line1.0h.BaseLine),

levels=design)

where group="cell_time_condition".

Is the correction by DMSO correctly done in my contrasts?

edger makecontrasts timecourse • 1.4k views

ADD COMMENT • link updated 7.5 years ago by Ryan C. Thompson ★ 7.9k • written 7.5 years ago by aec ▴ 90

score 0 · Answer 1 · 2016-11-08

0

Entering edit mode

Ryan C. Thompson ★ 7.9k

@ryan-c-thompson-5618

Last seen 8 months ago

Scripps Research, La Jolla, CA

This looks correct to me, but over-complicated. Notice that in each of your contrasts, the 2 cell_line1.0h.BaseLine terms cancel each other out.

ADD COMMENT • link 7.5 years ago Ryan C. Thompson ★ 7.9k

0

Entering edit mode

so how can I easily approach that?

ADD REPLY • link 7.5 years ago aec ▴ 90

1

Entering edit mode

I think you need to be more precise in defining what you want to test. You've written all your contrasts in the form (4h.treat - 0h) - (4h.control - 0h), which simplifies to 4h.treat - 4h.control. This is the difference between the 4h time point with and without the treatment. This is the only contrast that makes sense to me. Since the initial condition is the same for both control and treatment, the effect of the treatment at a given timepoint is represented by the differences between control and treatment at that time point. It might be useful to also test the contrast representing what happens to the control over 4 hours, i.e. 4h.control - 0h. You can then compare the direction of change relative to the 4h treatment effect to this to see if the treatment is reinforcing or opposing the changes that would normally occur in those 4 hours.

ADD REPLY • link 7.5 years ago Ryan C. Thompson ★ 7.9k

0

Entering edit mode

So apart from testing independently these two contrats:

4h.treat - 4h.control

4h.control - 0h

There is no other way to test the treatment effect (respect to time 0h) corrected by any effect of the DMSO (respect to time 0h) at the same time?

ADD REPLY • link 7.5 years ago aec ▴ 90

0

Entering edit mode

I think you're misunderstanding Ryan. It's not that there's no way to test for differences in the "corrected" effects. It's that there's no point in testing them, because mathematically, the time 0h baseline will cancel out; so you might as well directly test for differences between the 4h treatment and DMSO groups, i.e., 4h.treat - 4h.control.