Dear all,

Our current design seeks to find the impact of specific condition (compared to reference condition) on drug effect.

Two (2) conditions (including reference condition) and four (4) drug dosage (including vehicle) of one specific drug have been tested. Samples derived from the same mouse have been tested in each condition/drug combination. Multiple mouse have been used to create the final datasets.

      Sample condition SampleSource     drug
1   Sample_1         2     Mouse_01  vehicle
2   Sample_2         2     Mouse_01 Dose10uM
3   Sample_3         2     Mouse_01 Dose20uM
4   Sample_4         2     Mouse_01 Dose30uM
5   Sample_5         1     Mouse_01  vehicle
6   Sample_6         1     Mouse_01 Dose10uM
7   Sample_7         1     Mouse_01 Dose20uM
8   Sample_8         1     Mouse_01 Dose30uM
9   Sample_9         2     Mouse_02  vehicle
10 Sample_10         2     Mouse_02 Dose10uM
11 Sample_11         2     Mouse_02 Dose20uM
12 Sample_12         2     Mouse_02 Dose30uM
13 Sample_13         1     Mouse_02  vehicle
14 Sample_14         1     Mouse_02 Dose10uM
15 Sample_15         1     Mouse_02 Dose20uM
16 Sample_16         1     Mouse_02 Dose30uM
17 Sample_17         2     Mouse_03  vehicle
...

We would like to be certain that the design of our current analysis correctly manages the paired samples.

One objective is to detect genes which at one or more dosage points after vehicle show a condition effect.

Using LRT test, we have set the vehicle as the reference level for drug dosage (drug variable). We are using a SampleSource variable to represent the mouse from which the samples have been derived.

dds <- DESeqDataSetFromMatrix(countData = intMat, colData = meta, design = ~ SampleSource + condition + drug + condition:drug)
dds <- DESeq(dds, test="LRT", reduced = ~ SampleSource + condition + drug)
res<- results(dds, alpha = 0.05)

Question 1. Should the SampleSource variable be included in the reduced design in the DESeq() function?

Question 2. Is the order of condition and drug variables important?

Thanks, Astrid

"We would like to be certain that the design of our current analysis correctly manages the paired samples."

At this point, unfortunately I don't have time to provide statistical recommendations on the support site, I have to limit my extra time for software related questions. I'd therefore recommend consulting with a local statistician on design considerations.

I'll just mention that nuisance variables should be included in the reduced design (e.g. batch or patient, when blocking on those). And the order of factors is discussed in the vignette. It only matters if you call results with no arguments (which is maybe best to avoid with multi-factor designs, again, see vignette section on factor order).