nested mixed model in DESeq2

[Golsheed]

Hi,

I was wondering whether I could use a nested mixed model in DESeq2? Specifically, the model I have in mind has two fixed effects A and B, and a random effect C nested in B. I was wondering whether the following could work?
design = ~ A + B + (1| C %in% B)
I'd appreciate any help or comment.

Thanks,
Golsheed

[dpryan]

DESeq2 (as well as edgeR) doesn't support mixed-effect models.

[Golsheed]

Thanks for the help.

[Golsheed]

Could you please comment on the design formula, regardless of the DESeq2? I'm having a bit of a hard time verifying if it's correct. My model has two fixed effects A and B, and a random effect C nested in B. I was wondering whether the following could work?
formula = ~ A + B + (1| C %in% B)

Thanks!

[Golsheed]

Could you please comment on the design formula, regardless of the DESeq2? I'm having a bit of a hard time verifying if it's correct. My model has two fixed effects A and B, and a random effect C nested in B. I was wondering whether the following could work?
formula = ~ A + B + (1| C %in% B)

Thanks!

[Michael Love]

limma has functionality called duplicateCorrelation() for arrays or for samples. See this reference and the help file for the function in limma.

Smyth, G. K., Michaud, J., and Scott, H. (2005). The use of within-array replicate spots
for assessing differential expression in microarray experiments. Bioinformatics 21(9),
2067–2075.

http://bioinformatics.oxfordjournals.org/content/21/9/2067

Otherwise, this is more of a question for a mixed effects modeling package like lme4 -- which has glm support including the NB: glmer.nb() -- or for R-help.

[Golsheed]

Thanks a lot, Michael. I really appreciate your help. We're actually doing some part of our analysis with limma, so it's great to know that it supports mixed models. I'll look into it, as you suggested!

I actually used the glmer.nb() function to fit my data but am still not sure about the formula.

Suppose that the dependetn variable is Counts, and I have X1 and X2 as fixed effects, while X3 is a random effect and nested in X3. Specifically, X2 has two levels (L1 and L2), with 2 levels of X3 happening in (or belonging to) L1 and 8 levels of X3 belonging to L2. I'm guessing the naive way to model this is

Count ~ X1+ X2+ (1|X2/X3),

is that correct? But the random effect term (i.e., (1|X2/X3)) treats X2 as a random effect, by default, while I want X2 to be a fixed effect. So I was wondering whether the following would suffice to capture what I have in mind?
Count ~ X1+ X2+ (1|X3)

Thanks a bunch,
Golsheed

limma has functionality called duplicateCorrelation() for arrays or for samples. See this reference and the help file for the function in limma.

Smyth, G. K., Michaud, J., and Scott, H. (2005). The use of within-array replicate spots
for assessing differential expression in microarray experiments. Bioinformatics 21(9),
2067–2075.

http://bioinformatics.oxfordjournals.org/content/21/9/2067

Otherwise, this is more of a question for a mixed effects modeling package like lme4 -- which has glm support including the NB: glmer.nb() -- or for R-help.