design matrix for complicated experimental design in limma

Question

[Naresh D J]

Hi,

I am working with microarray data and its experimental design is bit strange. I have 20 samples (expression at 5 time points, at 1 hr with 3 replicates, 2 hr with 4 replicates, 3 hr with 3 replicates, 6hr with 6 replicates and 10hr with 4 replicates).

I want to find the DEGs at each time point vs rest of time points. For instance,

DEGs at 1 hr vs rest of all time points (2, 3, 6, 10)
DEGs at 2 hr vs rest of all time points (1, 3, 6, 10) 
DEGs at 3 hr vs rest of all time points (1, 2, 6, 10) etc..

My design matrix is as follows:

                complete hr1 hr2 hr3 hr6 hr10
1hr:rep1         1   1   0   0   0    0
1hr:rep2         1   1   0   0   0    0
1hr:rep3         1   1   0   0   0    0
2hr:rep1         1   0   1   0   0    0
2hr:rep2         1   0   1   0   0    0
2hr:rep3         1   0   1   0   0    0
2hr:rep4         1   0   1   0   0    0
3hr:rep1         1   0   0   1   0    0
3hr:rep2         1   0   0   1   0    0
3hr:rep3         1   0   0   1   0    0
6hr:rep1         1   0   0   0   1    0
6hr:rep2         1   0   0   0   1    0
6hr:rep3         1   0   0   0   1    0
6hr:rep4         1   0   0   0   1    0
6hr:rep5         1   0   0   0   1    0
6hr:rep6         1   0   0   0   1    0
10hr:rep1        1   0   0   0   0    1
10hr:rep2        1   0   0   0   0    1
10hr:rep3        1   0   0   0   0    1
10hr:rep4        1   0   0   0   0    1

Is this design matrix correct as per my DEGs comparison?

Moreover, I am getting this error/warning as Warning message:

" Partial NA coefficients for 20214 probe(s) ".

Can anyone help me what's wrong with my design matrix?

Thank you.

[andrew.j.skelton73]

Please can you add the code you used to create the design matrix, along with a phenotype table.

[Naresh D J]

Thank you. I got the answer from bioconductor. Link is provided here. https://support.bioconductor.org/p/85502/#85519

Answer 1

[Devon Ryan]

Either get rid of the 'complete' column or one of the time points. I recommend the former if you're going to use contrasts.

[Naresh D J]

Thank you Devon Ryan. Yes. you are right that I need to remove complete column and use the contrasts function. Answer is provided here. https://support.bioconductor.org/p/85502/#85519