DESeq2: How to estimate effects of one factor controlling for other factors?
1
0
Entering edit mode
mjd275 • 0
@cd933673
Last seen 18 days ago
France

Hi everyone,

I am trying to help a colleague who uses DESeq2. He has two factors for explaining the variability of genes and he wants to identify if there is an effect of each factor as well as a possible interaction. I am used to performing generalized linear models with different R packages where we can easily specify two factors with their interaction. Then there are statistical tests for determining if each factor has a global effect or not. In such approach, the effets of each factor are estimated controlling the effects of the other one. So I was wondering if, with DESeq2, we are able to estimate the effects of each factor controlling for the other ones?

Thanks for your help.

Best, Mary.

DESeq2 • 331 views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 5 days ago
United States

See the vignette where this is discussed (multi-factor designs).

DESeq2 is just like other R package, you specify a design formula.

ADD COMMENT
0
Entering edit mode

Yes. You're right specifying a design is not complicated with DESeq2. I am asking about how to obtain in one output, for example, p-values related to the significance of each factor. I have just seen some output only when focusing on a combination of factor's levels.

If we enter this formula: G1~ Sexe+Ttt +Sex:Ttt, may we obtain for example this type of output: Sex pvalue Ttt pvalue Sex:Ttt pvalue

Thanks.

ADD REPLY
0
Entering edit mode

Yes. You can get all of those things. Just not in a single table. Instead, you get three tables with the coefficients and p-values for each of the coefficients you care about (which you could presumably rbind together if that's important to you). Although in the context of an interaction you cannot interpret the main effects if the interaction is significant, so you have to be careful.

ADD REPLY
0
Entering edit mode

Thanks for your response. Just one last question: coefficients for each factor are obtaining by controlling for the other ones?

ADD REPLY
0
Entering edit mode

Yes. DESeq2 is simply fitting a GLM, just like if you were to fit a GLM to each gene separately. You interpret the results similarly.

ADD REPLY
0
Entering edit mode

As James said.

If you want to see the design matrix that is used, just call:

model.matrix(design(dds), colData(dds))

ADD REPLY
0
Entering edit mode

Thanks for your help !

ADD REPLY
0
Entering edit mode

Sorry a last question. I am going through a simple example and I have still one question about the global test associated to each factor.

I have two factors : dose ( with 3 levels D0.5/D1/D2) and ttt (with 2 levels). When using this R code results(dds, contrast =c("dose","D1","D2")) I get one p-value corresponding to "log2 fold change (MLE): dose D1 vs D2" . Is it possible to have only one pvalue for this factor? I guess I can play with the LRT test but I am just wondering if there is a way of computing a global test .

Thanks.

ADD REPLY
0
Entering edit mode

Sorry a last question. I am going through a simple example and I have still one question about the global test associated to each factor.

I have two factors : dose ( with 3 levels D0.5/D1/D2) and ttt (with 2 levels). When using this R code results(dds, contrast =c("dose","D1","D2")) I get one p-value corresponding to "log2 fold change (MLE): dose D1 vs D2" . Is it possible to have only one pvalue for this factor? I guess I can play with the LRT test but I am just wondering if there is a way of computing a global test .

Thanks.

ADD REPLY

Login before adding your answer.

Traffic: 855 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6