Hi all,

I have a question regarding contrasts as well as experiment design. My targets file looks like this:

SlideNumber    Name    FileName    Cy3    Cy5    info

1    A_1    A1.gpr    Ref    A1
2    A_2    A2.gpr    A2    Ref    dyeswap
3    B_1    B1.gpr    Ref    B1    
4    B_2    B2.gpr    B2    Ref    dyeswap
5    B_3    B3.gpr    Ref    B3    
6    C_1    C1.gpr    Ref    C1    
7    C_2    C2.gpr    C2    Ref    dyeswap
8    C_3    C3.gpr    Ref    C3    

I have 3 conditions A, B, C and Ref is supposed to be same in all experiments. Each condition should have 3 replicates, one being dyeswap but one replicate A3 failed, so we only have 2 replicates for that particular condition. Instead of cDNA we used genomic DNA (gDNA) for Reference but for the conditions it is always cDNA.

Can i use limma for the differential expression analysis of A vs. B and A vs. C and B vs.C. As i have only 2 replicates in the condition A, do the comparison or the normalization gives meaningful results?

Also can i use mean log fold changes of each condition A vs. Ref and B vs. Ref and use these values to calculate the fold changes A vs. B instead of contrasts as technically the Ref should be same and this also leads to A vs. B. Because if i do like this, and do using contrasts i get different log fold changes.

Thank you all

Having only 2 replicates for A is no problem at all.

However your targets file is incorrect. This is what you need:

> targets
   SlideNumber Name FileName Cy3 Cy5
A1           1  A_1   A1.gpr Ref   A
A2           2  A_2   A2.gpr   A Ref
B1           3  B_1   B1.gpr Ref   B
B2           4  B_2   B2.gpr   B Ref
B3           5  B_3   B3.gpr Ref   B
C1           6  C_1   C1.gpr Ref   C
C2           7  C_2   C2.gpr   C Ref
C3           8  C_3   C3.gpr Ref   C

Just do this:

> design <- modelMatrix(targets,ref="Ref")
Found unique target names:
 A B C Ref
> design <- cbind(Dye=1,design)
> fit <- lmFit(MA, design)

Then take contrasts between A, B and C. It's all pretty straightforward -- see the limma guide.

I think you are misunderstanding how contrasts and fold changes work. limma does compute the contrasts by comparing each treatment back to the common reference.