I have measurements from 12 mice, grouped in two conditions. I each mouse I have measurements from 4 tissues. The design is not balanced, 5 mice in condition1 and 7 in condition2.
After reading the bioconductor edgeR manual I have set up the following model:
design <- model.matrix(~ Condition + Tissue + Condition:Mouse + Condition:Tissue)
I have then manually removed terms with no observations.
design <- design[, -which(colnames(design) %in% c("Condition1:Mouse6", "Condition1:Mouse7"))]
I can fit the model, but when I set up contrasts I get confused. I want to get the main effect of being in condition2, the main effect of each tissue and the interaction effect for each tissue.
colnames(design)  "(Intercept)" "Condition2" "Tissue2"  "Tissue3" "Tissue4" "Condition1:Mouse2"  "Condition2:Mouse2" "Condition1:Mouse3" "Condition2:Mouse3"  "Condition1:Mouse4" "Condition2:Mouse4" "Condition1:Mouse5"  "Condition2:Mouse5" "Condition2:Mouse6" "Condition2:Mouse7"  "Condition2:Tissue2" "Condition2:Tissue3" "Condition2:Tissue4"
To get the main effect of condition I use the contrast
Comparing e.g. Tissue 1 and 2 or 2 and 3 I do like this
But say I want to get the difference between condition 1 and 2 in tissue 1. Given how the model matrix is built tissue 1 is not present. Can I do a comparison like:
And how about an anova like test for changes across all the tissue:condition interaction terms. The contrast
Would compare only tissue2, 3 and 4, and with what?
Many questions here. Quick recap. 1. Is it OK to remove empty terms in an unbalanced design? 2. How do I find the effect of the level in the tissues category that is missing in the model matrix. 3. How do I find changes happening across all levels of the interaction terms (related to the missing tissue1 problem).
A link to a page with a thorough R model matrix/linear model guide would also be appreciated.