Limma Multiple Group Analysis

Question

[piyu.riya]

Hi, I am using limma to do a multiple group analysis using Affymatrix data. I am trying to perform analysis for 12 samples which are from control (C) and disease (D) group and these are either treated with drug (S) or not (N). The replicas are not consistence across the group. I run the following code:

types <- factor(types, levels=c("C.S","C.N","D.S","D.N"))
design <- model.matrix(~ 0+types)
colnames(design) <- levels(types)
contrast.matrix <- makeContrasts(C.SvsN=C.S-C.N,
+ D.SvsN=D.S-D.N,
+ Diff=(D.S-D.N)-(C.S-C.N),
+ levels=design)

But when I run the below code

fit <- lmFit(exprs(eset),design)

I get the error message:

"Coefficients not estimable: C.N D.N "
"Error in lm.fit(design, t(M)) : incompatible dimensions"

I will really appreciate if someone can help me to understand this error and help me to debug this problem.

Thanks, Riya

[Devon Ryan]

A few comments might prove helpful. Firstly, can you post the model matrix? The error message suggests that you're using either a not-so-great experimental setup or (more likely) that there's simply a better model to use that doesn't have the non-estimable coefficients problem. Secondly, have you considered simply using a factorial design? I suspect this is the real answer to your problem. Have a look at section 8.5 in the limma users guide.

Answer 1

[Gordon Smyth]

The first error message is telling you that there aren't any samples with C.N or D.N. In other words, there are no cases not drug treated.

The second message is telling you that the number of rows of the design matrix doesn't match the number of columns in your data.

It seems you have not defined the vector 'types' correctly. Perhaps you have just left some of the cases out so that the design matrix has two few rows.