Contrast treated samples against both scramble and untreated controls
1
0
Entering edit mode
@oliverziff-22402
Last seen 3 days ago
United Kingdom

Hi there, I am using DESeq2 to analyse RNAseq from siRNA treated samples and 2 controls (Scramble and Untreated). Each treatment has 4 cell lines:

   treatment cellline group
<fct>     <fct>    <fct>
1 Untreated  1       Control
2 Scramble   1       Control
3 Knockdown  1       Knockdown
4 Untreated  2       Control
5 Scramble   2       Control
6 Knockdown  2       Knockdown
7 Untreated  3       Control
8 Scramble   3       Control
9 Knockdown  3       Knockdown
10 Untreated  4       Control
11 Scramble   4       Control
12 Knockdown  4       Knockdown


I would like to contrast Knockdown versus all Control samples but remove effects due to (i) cell line and (ii) scramble. I would like to use the most optimal design and contrast to achieve this.

The straight forward options are to either lump together Untreated and Scramble into a single Control group:

design = ~ cellline + group
results(dds, contrast=c("group", "Knockdown", "Control"))


or ignore Untreated samples all together:

design = ~ cellline + treatment
results(dds, contrast=c("group", "Knockdown", "Scramble"))


Neither are optimal and I wondered whether there is a better option that incorporates Scramble treatment into the design but avoid the error: model matrix is not full rank, so the model cannot be fit as specified. For example, could i specify a likelihood ratio rest with the full model = ~ cellline + treatment + group and reduced = ~ cellline + treatment?

Many thanks! Oliver

design contrast DESeq2 • 183 views
1
Entering edit mode
@mikelove
Last seen 1 day ago
United States

You can do ~cellline + treatment. Now you in actuality have three groups of samples aside from the cell-line differences. When you want to compare KD vs the two controls, do you want to just average the difference between KD vs scramble and KD vs untreated? You can do that with a numeric contrast vector.

0
Entering edit mode

Thank you Michael Love !

Yes, averaging the difference between Knockdown vs scramble and Knockdown vs untreated is a much better solution. So using design = ~ cellline + treatment and then:

mod_mat <- model.matrix(design(dds), colData(dds))
Knockdown <- colMeans(mod_mat[dds$treatment == "Knockdown", ]) Control <- colMeans(mod_mat[dds$treatment %in% c("Scramble", "Untreated"),]) # vector of coefficients for the average of Scramble and Untreated
kd_vs_ctrl <- results(dds, contrast = Knockdown - Control)


Thanks so much!

1
Entering edit mode

Yes, for a typical model.matrix setup it should give c(0,-.5,1)