DESeq2: "model matrix is not full rank"

Question

[Jalil Sharif]

Hello, I have the following variables for my deseq2 analysis but I am getting the error model matrix is not full rank.

condition   type
control 1
control 1
control 1
ko  2
ko  2
ko  2
ko  3
ko  3
ko  3
ko  4
ko  4
ko  4

I have four cell types, 1 wild-type and three different KO's for the same gene.

[Asaf]

Can you add the design? I assume it's because you have both condition and type in your design.

Answer 1

[ATpoint]

You cannot include both condition and type into the design, because control and type-1 are nested. You can either do ~condition, but this assumes that all KO are the same group. If they're heterogeneous then it will drive up dispersion and potentially get few DEGs. Alternatively, do ~type (make sure it's a factor, not a number), and then you can do pairwise comparisons as you like between types.

Answer 2

[Asaf]

You basically have the situation described here. The solution is in there.