DESeq2: Multi-factor designs

[sindrle]

With great help from dpryan I have come a long way, but now Im stuck again. I need to be able to calculated both within and between samples. I have two groups at two time points, the data is paired within samples, but not between:

sampleFiles <- list.files(path="/Volumes/timemachine/HTseq_DEseq2",pattern="*.txt");
status <- factor(c(rep("Healthy",26), rep("Diabetic",22)))
timepoints = as.factor(c(1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,2,2,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2));
sampleTable <- data.frame(sampleName = sampleFiles, fileName = sampleFiles, status=status, timepoints=timepoints);
directory <- c("/Volumes/timemachine/HTseq_DEseq2/");
des <- formula(~timepoints+status);
ddsHTSeq <- DESeqDataSetFromHTSeqCount(sampleTable = sampleTable, directory = directory, design=des);
ddsHTSeq

These commands all work, but they are incorrect due to not taken the paired data in account.
So I looked here: page 49; http://www.bioconductor.org/packages...usersguide.pdf

So I tried instead;

Treat <- factor(paste(sampleTable$status,sampleTable$timepoints,sep=""));
design <- model.matrix(~0+Treat);
colnames(design) <- levels(Treat);

> design
Diabetic1 Diabetic2 Healthy1 Healthy2
1 0 0 1 0
2 0 0 1 0
3 0 0 1 0
4 0 0 1 0
5 0 0 1 0
6 0 0 1 0
7 0 0 1 0
8 0 0 1 0
9 0 0 1 0
10 0 0 1 0
11 0 0 1 0
12 0 0 1 0
13 0 0 1 0
14 0 0 0 1
15 0 0 0 1
16 0 0 0 1
17 0 0 0 1
18 0 0 0 1
19 0 0 0 1
20 0 0 0 1
21 0 0 0 1
22 0 0 0 1
23 0 0 0 1
24 0 0 0 1
25 0 0 0 1
26 0 0 0 1
27 1 0 0 0
28 1 0 0 0
29 1 0 0 0
30 1 0 0 0
31 1 0 0 0
32 1 0 0 0
33 1 0 0 0
34 1 0 0 0
35 1 0 0 0
36 1 0 0 0
37 1 0 0 0
38 0 1 0 0
39 0 1 0 0
40 0 1 0 0
41 0 1 0 0
42 0 1 0 0
43 0 1 0 0
44 0 1 0 0
45 0 1 0 0
46 0 1 0 0
47 0 1 0 0
48 0 1 0 0
attr(,"assign")
[1] 1 1 1 1
attr(,"contrasts")
attr(,"contrasts")$Treat
[1] "contr.treatment"

However my "old" design, the "des", looks like this:

> des
~timepoints + status

Running with "new" design:
ddsHTSeq <- DESeqDataSetFromHTSeqCount(sampleTable = sampleTable, directory = directory, design=design)

Gives error:

Error in formula.default(design) : invalid formula

Not to suprising.. But Im stuck, what to do? Am I totally of?

Thanks!

[jingerlu]

what colData(ddsHTSeq) gives ?

[sindrle]

The answer is, thanks to dpryan:

status <- factor(c(rep("Healthy",26), rep("Diabetic",22)), levels=c("Healthy", "Diabetic"));
timepoints = as.factor(c(1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,2,2,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2));
patients <- factor(c(04,21,53,55,61,62,67,73,76,77,79,04,21,53,55,61,62,67,73,76,77,79, 01,08,13,70,71,72,75,81,82,83,86,87,88,01,08,13,70,71,72,75,81,82,83,86,87,88));
sampleTable <- data.frame(sampleName = sampleFiles, fileName = sampleFiles, status=status, timepoints=timepoints, patients=patients);

and

des <- formula(~patients + timepoints*status);

[jingerlu]

I suppose this could get the pair combination you like :
Treat <- factor(paste(sampleTable$status,sampleTable$timepoints,sep=""));
design <- formula(~ 0 + Treat)

but I am not sure if the following DESeq() or nbinomWaldTest() or nbionomLRT() function can work with this design formula . I am so temped to switch to edgeR , which has nicer tutorial and examples to follow.

[sindrle]

So you mean I dont have paried test as I have wrote it?

[Simon Anders]

With the "~ patient + ...", you get paired tests. You currently test for interaction between time and status, i.e., you will get gene for which the amount of change between the time points differs significantly between healthy and diabetic subjects.

[sindrle]

Thanks!
That what I thought, but is that the only thing tested?

I get alot of results, of which I have chosen three:

statusResults <- results(dds, "status_Healthy_vs_Diabetic");
timepointsResults <- results(dds, "timepoints_2_vs_1");
statusTreatmentResults <- results(dds, "timepoints2.statusHealthy");

So you described the "timepoints2.statusHealthy" result, right?

But am I correct to use "status_Healthy_vs_Diabetic" to look for differences between the groups (disregarding time), and vise versa for "timepoints_2_vs_1"?

Thanks!

[dpryan]

Treat <- factor(paste(sampleTable$status,sampleTable$timepoints,sep=""));
design <- formula(~ 0 + Treat)

Will create 4 groups: "Diabetic1", "Diabetic2", "Normal1", and "Normal2". So no, that won't keep the pairing. This is a similar design to what cuffdiff would do if you input the files as 4 groups. Aside from the pairing issue, some people prefer this since it's easier to directly compare groups, which is what they want to do. It's mostly a matter of the question you really want to ask.

[sindrle]

So to compare Cuffdiff2 with DESeq2 I could run that design (since I have already done that on Cuffdiff 2), but as you said its not my actual question of interest, thus leading me to DEseq2 in the first place.

[Simon Anders]

But am I correct to use "status_Healthy_vs_Diabetic" to look for differences between the groups (disregarding time), and vise versa for "timepoints_2_vs_1"?

For those genes for which the interaction is not significant: Yes.

For the others: not quite because you cannot disregard time because the difference depends on the time point.

If you want to average over the two time points, you can add half of the interaction effect to the status main effect (or, vice versa, add half of the interaction to the time-point main effect to get the average over disease states).

[sindrle]

What if time had the same effect on both groups?
Or if groups are the same, but both changed in time.

Could you then do as I said?