I am trying to run multi factor comparison by DESeq2 and keep getting errors. My colData is a data frame, named condition, that is written like the table below.

period type

1 A 3

2 A 3

3 B 1

4 B 2

5 C 1

6 C 2

7 D 1

8 D 2

I want to compair any 2 periods considering type, specifically, A(3,3) vs B(1,2), B(1,2) vs C(1,2)...and C(1,2) vs D(1,2). And I also want to compair samples belonging to type1 and samples belonging to type2 regardless of period. I am a beginner to use DESeq2, so may I trouble you to tell me how to give a correct design and the full commands to implement my goal. Thank you very much.

Period A and type 3 are always the same, so you can't use ~period+type as a model and hope for it to work (I imagine that this is the error you're getting). There are a few steps in DESeq2 where you can manually input a model matrix, so you might have luck going that route. I'm not going to give you the exact commands, you should at least attempt to figure those out yourself, but the general idea is:

m <- model.matrix(~period+type, colData(dds))

Remove the column for type.3 (probably the last one). Your model matrix should now not be rank deficient.

See post #29 here from Michael Love where he shows how to manually specify a model matrix in some of the DESeq2 function.

You can then scroll down on to post #34 and #35 on the same page where examples are given for performing contrasts. Comparing type 1 vs. type 2 can be done simply with the coef= parameter in the results() function. See the DESeq2 vignette.

BTW, you'd probably get faster feedback if you'd put DESeq2 in the title of your post (the package authors are quite quick at answering questions...but they need to know they're there).

hi Emily, Devon's right about the workaround. (in 4, use name instead of coef.) I am planning to build more support for custom design matrices this devel cycle, for the cases like these where handing a formula to model.matrix() is not sufficient.

Also, here is another post where I talk about the v1.6 workaround:

I find my design has something wrong. Now I have adjusted my design and it is like the table below:

sample period sex individual

X2 stageP F 1
X9 stageP F 2
X16 stageP F 3
X23 stageP F 4
X3 stageP M 1
X10 stageP M 2
X17 stageP M 3
X24 stageP M 4
X4 stageG F 1
X11 stageG F 2
X18 stageG F 3
X25 stageG F 4
X5 stageG M 1
X12 stageG M 2
X19 stageG M 3
X26 stageG M 4
X6 stageM F 1
X13 stageM F 2
X20 stageM F 3
X27 stageM F 4
X7 stageM M 1
X14 stageM M 2
X21 stageM M 3
X28 stageM M 4

individuals with the same number in the same stages are paired.

I want to make comparisions below:

(1)results between different stages within one sex

(2)in the three stages respectively, results between the paired F and M

I'm sorry that I can not understand the post that you have given. Would you like to give me the formular=~?+?:?... that I should use to get my expected calculation? And may I have the content of the contrast that I would use to extract my results? Holding up too much of your time, I am so sorry and I really appreciate your selfless help.Thank you very much.

Just a note, it would have been better if you could have just posted the real table to start, as it seems the previous answers from Devon and I don't really apply to this table.

"individuals with the same number in the same stages are paired"

I'm confused how individuals with the same number in the same stages can be paired, because they have different sex. In what way do you mean they are paired?

"(2)in the three stages respectively, results between the paired F and M"

just to be clear, by this do you mean, the F vs M effect for each stage, taking into account the individual effect?

You can use a design: ~ period + period:individual + period:sex

which will allow you to test for period specific differences due to sex, while controlling for individual differences, where individuals with the same number in the same stages are paired.

The meaning of this last contrast is the sum of the P vs G effect for F (the base level) and the contrast of P vs G for M. Interactions work by adding extra effects to the base level effect.

I am sorry that I do not explain my idea correctly.I now describe them more carefully, hoping that you can get it exactly. May you spend some time read them again?

my purposes are:

(1)make paired t-test: I want to compute F vs M effect for each stage, taking into account the individual effect, because the F part and M part with the same individual numbers are come from one organism.

(2)make ANOVA: I want to compute stage effect for each sex, not taking into account the individual effect.

Thank you for your constant help and patience. I would try it again following your guide. And if there is anything else that I can not solve, sincerely hope your comment. Thank you very much.

Thanks Devon.

hi Emily, Devon's right about the workaround. (in 4, use

`name`

instead of`coef`

.) I am planning to build more support for custom design matrices this devel cycle, for the cases like these where handing a formula to model.matrix() is not sufficient.Also, here is another post where I talk about the v1.6 workaround:

C: DESeq2 paired and multi factor comparison