Differential Expression Analysis of Two Groups Time Series Data with Limma

Question

[hkarakurt]

Hello, I have a microarray gene expression data set consists 30 samples for two groups (Wild Type and Mutant) with certain time points. I want to do differential expression analysis between conditions but I want my analyses to consider time points too. My data looks like this: Sample Condition Time Point 1 WT 32 2 WT 32 3 WT 32 4 WT 42 5 WT 42 6 WT 42 7 WT 49 8 WT 49 9 WT 49 10 WT 56 11 WT 56 12 WT 56 13 WT 66 14 WT 66 15 WT 66 16 MU 32 17 MU 32 18 MU 32 19 MU 42 20 MU 42 21 MU 42 22 MU 49 23 MU 49 24 MU 49 25 MU 56 26 MU 56 27 MU 56 28 MU 66 29 MU 66 30 MU 66

And my codes are here:

condition <- c(rep("WT",15) , rep("MU",15))
time_point <- c(rep("32h",3),rep("42h",3),rep("49h",3),rep("56h",3),rep("66h",3),rep("32h",3),rep("42h",3),rep("49h",3),rep("56h",3),rep("66h",3))

exp <- paste(condition , time_point , sep = "_")

exp <- as.factor(exp)

design = model.matrix(~exp)

fit = lmFit(data , design)

fit <- eBayes(fit)

When I construct my design matrix, first column is intercept (full of 1s) and I cannot see WT_32h in the matrix. Is there a better way to do it of should I trust my results with these codes?

Thank you in advance

[h.mon]

You can trust the results, but you have to be careful about the contrasts to get what you want. If you specify your design as

design = model.matrix( ~ 0 + exp )

The matrix will have one column for treatment and won't include an intercept. You may want to read section "9.6 Time Course Experiments" from the limma Users Guide.

[hkarakurt]

Thank you so much for your answer. I tried your advice but there is a problem apparently. When I use "design = model.matrix( ~ 0 + exp )" about 95% of the genes are signficantly changed and it looks wrong. With my codes it was about 30%.

What is the disadvantage of using "~exp" instead of "~0 + exp" in that case?

Thank you.