DEseq2 with different designs matrix

Question

[QX]

Hi all,

I have DEseq2 questions on designs matrix:

samples rep day
A   1   3
A   2   3
B   1   3
B   2   3
C   1   3
C   2   4
D   1   4
D   2   4
F   1   4
F   2   4
G   1   5
G   2   5
H   1   5
H   2   5

and command

dds_all <- DESeqDataSetFromMatrix(countData = count_table,
                                  colData = designs,
                                  design = ~ samples)

If I decide to remove some samples from the desgin and count_table, let say G and H, and do the pair-wise comparison between A and B, do I need to make the new count table and design matrix without sample G and H?

And, if on other analysis, I only have interest on pair-wise comparison between day 3 and day 4, do I need to perform new DEseq2 run that contain only rows for day 3 and 4 (A B C D F), or Can I used the exact same design matrix and do the results("day","3","4")? if I dont remove the sample from day 5, will it affect the scale factor of DEseq2?

[BioinfGuru]

I suggest going through this workflow to get a better understanding.

Answer 1

[swbarnes2]

You can include all the samples in one object, and do any or all pairwise comparisons you want. This is in general the better way to handle it.

Drop replicate from your design. You can't use it. You've also got a problem with sample/day confounding.

[QX]

Hi, thank for your answer. I don't drop replicate, I just simply do not have interest in some samples and intend to remove them from the whole analysis. Can you explain more about the 'problem with sample/day confounding'?