Question

edgeR design matrix with multiple cell types and combination treatments

0

Entering edit mode

giBio • 0

@7a253a61

Last seen 2.4 years ago

Germany

Hello everyone,

I have the experimental data involving different cell lines treated with either DMSO (vehicle control) or 2 individual drugs as well as their combination - each sample in 3 replicates. I'm interested in comparing DEG after treatment with individual drugs and their combination within and across cell lines. I am going through various guides and vignettes, but still finding the matrix design very complex. I have come with the following design:

Table: sampleInfo

cell_line	cell_type	cell_group	treatment	treat1	treat2	replicate
A	cellLine	A_cellLine	DMSO	NO	NO	1
A	cellLine	A_cellLine	DRUG1	YES	NO	1
A	cellLine	A_cellLine	DRUG2	NO	YES	1
A	cellLine	A_cellLine	COMBI	YES	YES	1
A	cellLine	A_cellLine	DMSO	NO	NO	2
A	cellLine	A_cellLine	DRUG1	YES	NO	2
A	cellLine	A_cellLine	DRUG2	NO	YES	2
A	cellLine	A_cellLine	COMBI	YES	YES	2
A	cellLine	A_cellLine	DMSO	NO	NO	3
A	cellLine	A_cellLine	DRUG1	YES	NO	3
A	cellLine	A_cellLine	DRUG2	NO	YES	3
A	cellLine	A_cellLine	COMBI	YES	YES	3
B	cellLine	B_cellLine	DMSO	NO	NO	1
B	cellLine	B_cellLine	DRUG1	YES	NO	1
B	cellLine	B_cellLine	DRUG2	NO	YES	1
B	cellLine	B_cellLine	COMBI	YES	YES	1
B	cellLine	B_cellLine	DMSO	NO	NO	2
B	cellLine	B_cellLine	DRUG1	YES	NO	2
B	cellLine	B_cellLine	DRUG2	NO	YES	2
B	cellLine	B_cellLine	COMBI	YES	YES	2
B	cellLine	B_cellLine	DMSO	NO	NO	3
B	cellLine	B_cellLine	DRUG1	YES	NO	3
B	cellLine	B_cellLine	DRUG2	NO	YES	3
B	cellLine	B_cellLine	COMBI	YES	YES	3
C	primaryCells	C_primaryCells	DMSO	NO	NO	1
C	primaryCells	C_primaryCells	DRUG1	YES	NO	1
C	primaryCells	C_primaryCells	DRUG2	NO	YES	1
C	primaryCells	C_primaryCells	COMBI	YES	YES	1
C	primaryCells	C_primaryCells	DMSO	NO	NO	2
C	primaryCells	C_primaryCells	DRUG1	YES	NO	2
C	primaryCells	C_primaryCells	DRUG2	NO	YES	2
C	primaryCells	C_primaryCells	COMBI	YES	YES	2
C	primaryCells	C_primaryCells	DMSO	NO	NO	3
C	primaryCells	C_primaryCells	DRUG1	YES	NO	3
C	primaryCells	C_primaryCells	DRUG2	NO	YES	3
C	primaryCells	C_primaryCells	COMBI	YES	YES	3
D	primaryCells	D_primaryCells	DMSO	NO	NO	1
D	primaryCells	D_primaryCells	DRUG1	YES	NO	1
D	primaryCells	D_primaryCells	DRUG2	NO	YES	1
D	primaryCells	D_primaryCells	COMBI	YES	YES	1
D	primaryCells	D_primaryCells	DMSO	NO	NO	2
D	primaryCells	D_primaryCells	DRUG1	YES	NO	2
D	primaryCells	D_primaryCells	DRUG2	NO	YES	2
D	primaryCells	D_primaryCells	COMBI	YES	YES	2
D	primaryCells	D_primaryCells	DMSO	NO	NO	3
D	primaryCells	D_primaryCells	DRUG1	YES	NO	3
D	primaryCells	D_primaryCells	DRUG2	NO	YES	3
D	primaryCells	D_primaryCells	COMBI	YES	YES	3

Matrix design:


design <- model.matrix(~0+cell_group+(treat1*treat2), data = sampleInfo)
design
   cell_groupA_cellLine cell_groupB_cellLine cell_groupC_primaryCells cell_groupD_primaryCells treat1YES treat2YES
1                     1                    0                        0                        0         0         0
2                     1                    0                        0                        0         1         0
3                     1                    0                        0                        0         0         1
4                     1                    0                        0                        0         1         1
5                     1                    0                        0                        0         0         0
6                     1                    0                        0                        0         1         0
7                     1                    0                        0                        0         0         1
8                     1                    0                        0                        0         1         1
9                     1                    0                        0                        0         0         0
10                    1                    0                        0                        0         1         0
11                    1                    0                        0                        0         0         1
12                    1                    0                        0                        0         1         1
13                    0                    1                        0                        0         0         0
14                    0                    1                        0                        0         1         0
15                    0                    1                        0                        0         0         1
16                    0                    1                        0                        0         1         1
17                    0                    1                        0                        0         0         0
18                    0                    1                        0                        0         1         0
19                    0                    1                        0                        0         0         1
20                    0                    1                        0                        0         1         1
21                    0                    1                        0                        0         0         0
22                    0                    1                        0                        0         1         0
23                    0                    1                        0                        0         0         1
24                    0                    1                        0                        0         1         1
25                    0                    0                        1                        0         0         0
26                    0                    0                        1                        0         1         0
27                    0                    0                        1                        0         0         1
28                    0                    0                        1                        0         1         1
29                    0                    0                        1                        0         0         0
30                    0                    0                        1                        0         1         0
31                    0                    0                        1                        0         0         1
32                    0                    0                        1                        0         1         1
33                    0                    0                        1                        0         0         0
34                    0                    0                        1                        0         1         0
35                    0                    0                        1                        0         0         1
36                    0                    0                        1                        0         1         1
37                    0                    0                        0                        1         0         0
38                    0                    0                        0                        1         1         0
39                    0                    0                        0                        1         0         1
40                    0                    0                        0                        1         1         1
41                    0                    0                        0                        1         0         0
42                    0                    0                        0                        1         1         0
43                    0                    0                        0                        1         0         1
44                    0                    0                        0                        1         1         1
45                    0                    0                        0                        1         0         0
46                    0                    0                        0                        1         1         0
47                    0                    0                        0                        1         0         1
48                    0                    0                        0                        1         1         1
   treat1YES:treat2YES
1                    0
2                    0
3                    0
4                    1
5                    0
6                    0
7                    0
8                    1
9                    0
10                   0
11                   0
12                   1
13                   0
14                   0
15                   0
16                   1
17                   0
18                   0
19                   0
20                   1
21                   0
22                   0
23                   0
24                   1
25                   0
26                   0
27                   0
28                   1
29                   0
30                   0
31                   0
32                   1
33                   0
34                   0
35                   0
36                   1
37                   0
38                   0
39                   0
40                   1
41                   0
42                   0
43                   0
44                   1
45                   0
46                   0
47                   0
48                   1
attr(,"assign")
[1] 1 1 1 1 2 3 4
attr(,"contrasts")
attr(,"contrasts")$cell_group
[1] "contr.treatment"

attr(,"contrasts")$treat1
[1] "contr.treatment"

attr(,"contrasts")$treat2
[1] "contr.treatment"


# Data fitting
y <- calcNormFactors(y)
y <- estimateDisp(y, design)
fit <- glmFit(y, design)

However, I am struggling with the contrast design to incorporate the drug combination information. Any help would be very appreciated.

Thank you!

DifferentialExpression celltype treatment edgeR limma • 1.9k views

ADD COMMENT • link 2.5 years ago giBio • 0

0

Entering edit mode

Do you want drug results separately for each cell line? Or do you want drug effects averaged over all the cell lines? I'd have expected you would want the former but your design matrix is doing the latter.

What is different about the different replicates? Are samples in the same replicate more similar than samples in different replicates?

ADD REPLY • link 2.5 years ago Gordon Smyth 51k

0

Entering edit mode

My primary goal is to identify drug effects separately for each cell line. But I'm also interested in the overall effects of drugs across all cell lines. Replicates represent biological replicates: 3 different cell passages following similar treatments. Hence samples from the same replicate are not similar compared to samples in different replicates. Rather, samples with particular treatment in a specific cell line are (expectedly) similar across different replicates.

ADD REPLY • link 2.5 years ago giBio • 0

0

Entering edit mode

Then that inter-replicate correlation should be included in the analysis.

ADD REPLY • link 2.5 years ago Gordon Smyth 51k

0

Entering edit mode

Thank you for your very important advice. I do indeed check the MDS plot for inter-replicate correlation.

ADD REPLY • link 2.5 years ago giBio • 0

score 1 · Answer 1 · 2022-06-13

1

Entering edit mode

Gordon Smyth 51k

@gordon-smyth

Last seen 1 hour ago

WEHI, Melbourne, Australia

The standard approach is to analyse the study as a series of groups, an approach which is both comprehensive and straightforward:

Group <- paste(cell_line, treatment, sep=".")
Group <- factor(Group)
design <- model.matrix(~0+Group)
colnames(design) <- levels(Group)

You can then form contrasts to make any comparison you want and the meaning of the contrasts becomes explicit.

Since your experiment is conducted as a series of biological replicates, you should also include that error structure in the analysis. If you were using limma rather than edgeR, you could set up cell-specific biological replicates

BiolRep <- paste(cell_line, replicate, sep=".")

and then use duplicateCorrelation wth block = BiolRep. In edgeR, it's a bit more complicated. In edgeR, you need to add cell line specific replicate effects to the design matrix.

ADD COMMENT • link 2.5 years ago Gordon Smyth 51k

0

Entering edit mode

Thank you very much for your suggestion. I think this is the comparison using a single factor. Following this, I designed the contrast as below:

y <- calcNormFactors(y)
y <- estimateDisp(y, design)
fit <- glmFit(y,design)

# To calculate the effect of individual drug in a particular cell line (against vehicle control DMSO)
A.DRUG1.deg <- glmLRT(fit, contrast = makeContrasts(A.DRUG1-A.DMSO, levels=colnames(design))) # and so on for other pairwise contrasts

# To calculate the effect of drug combination compared to individual drugs in a particular cell line (against vehicle control DMSO)
A.deg <- glmLRT(fit, contrast = makeContrasts((A.COMBI-A.DMSO)-(A.CYT-A.DMSO)-(A.DRN-A.DMSO),
                                          levels=colnames(design)))

Do you think this would be the right approach to understand the effect of the combination compared to individual drugs?

(Reference: Section: 6.2 Interaction using a single factor model)

ADD REPLY • link 2.5 years ago giBio • 0

0

Entering edit mode

I cannot tell you which contrasts to form because that is a biological question rather than a statistical question and only you can know what scientific questions are relevant to your study.

The contrast you have chosen is the classical interaction contrast, which tests whether the combination effect is equal to the sum of the individual drug effects. In other words, it tests for a synergistic combination effect. In my opinion, the classical interaction contrast is seldom of interest in these biological contexts, despite the long history interaction theory in statistics. Why would you expect the combination effect to be the sum of the individual effects? If it was me, I would start out with much more modest aims, for example by establishing whether any of the drugs have an effect, and whether the combination effects seem bigger than either of the individual effects.

ADD REPLY • link 2.5 years ago Gordon Smyth 51k

0

Entering edit mode

Great advice. From the experimental data, we know that the combination is synergistic. But this doesn't mean that I should expect the statistical synergism at the level of DEG. I will consider the individual treatment groups to identify DEG and then compare those groups in downstream analyses. Thank you very much for your help!

ADD REPLY • link 2.5 years ago giBio • 0