Limma multifactor experiment design matrix

[Lugalbanda]

Hi, so I am very confused as to how I can make a multifactor design for limma RNA differential expression. I've read other posts in different forums and none of the answers are very explicit (to me) as to how to build a multifactor design

For examples, say I have p1N, p1T, p2N, and p2T, where p# is patient number and N/T is normal tumor.

I can build a single factor design matrix like this:

------Normal------Tumor
1 ------ 1 --------- 0
2 ------ 0 --------- 1
3 ------ 1 --------- 0
4 ------ 0 --------- 1

I am completely lost as to how I can build a numeric matrix with multiple factors, which in this case, the factors would be patient number and N/T.

[dpryan]

d = data.frame(patient=factor(c("p1", "p1", "p2", "p2", ...)),
               condition=factor(c("N", "T", "N", "T", ...)))
design = model.matrix(~patient + condition, d)

N.B., I've likely made a typo somewhere, but that's the gist.

[Lugalbanda]

Oh, I thought I had to manually build a numeric matrix for the design matrix. Thanks. I have a followup question. I understand that the intercept column that is resulted is used for pair wise comparison. My question is when to use:

design = model.matrix(~0+patient + condition, d)
as opposed to this
design = model.matrix(~patient + condition, d)

Ok I found this from another post:

EDIT:
"This [design = model.matrix(~0+patient + condition, d)] (a cell means model) simply computes the mean expression for each group, and then you have to make all contrasts explicitly. The [design = model.matrix(~patient + condition, d)] has an implicit contrast (everything is a comparison to control), so you have to make contrasts for some comparisons, but not for others."

from https://support.bioconductor.org/p/57268/


So is it safe to say this only matters if I am doing a contrast matrix comparison?

[dpryan]

Correct, the "0+" is only needed if you want to specify contrasts, which aren't needed to answer your question.