Question: DESeq2--compare two log2-transformed fold changes in paired and multi factor sample
0
3 months ago by
tntntntntn0 wrote:

Hi,

I have RNA-seq dataset from two population (1and 2) including some patients and two conditions (B and A, before and after treatment) for each person.

My interest is whether there is a difference between population in the effect of treatment. In other words, I would like to compare whether log2(1A/1B) equals log2(2A/2B) or not.

I read DESEQ2--compare two log2-transformed fold changes, and I am generating a model using interaction terms. The difference is that there are some patients in each group and I have to consider paring each of them.

My sample table is like this:

population patient condition

1 p1 B

1 p2 B

1 p3 B

1 p4 B

1 p5 B

1 p1 A

1 p2 A

1 p3 A

1 p4 A

1 p5 A

2 p6 B

2 p7 B

2 p8 B

2 p6 A

2 p7 A

2 p8 A

Since patient is nested in population, I set patient.nested.

population patient condition patient.nested

1 p1 B p1

1 p2 B p2

1 p3 B p3

1 p4 B p4

1 p5 B p5

1 p1 A p1

1 p2 A p2

1 p3 A p3

1 p4 A p4

1 p5 A p5

2 p6 B p1

2 p7 B p2

2 p8 B p3

2 p6 A p1

2 p7 A p2

2 p8 A p3

Then, I would like to create a modelMatrix and remove columns of levels without samples. However, I cannot figure out what model is appropriate for my purpose. Does someone havs suggestions on how to design the model in this situation in DESeq2?

Regards,

Tatsuhiko

deseq2 • 125 views
modified 3 months ago by mikhael.manurung200 • written 3 months ago by tntntntntn0
Answer: DESeq2--compare two log2-transformed fold changes in paired and multi factor sam
1
3 months ago by
Netherlands
mikhael.manurung200 wrote:

Due to the paired nature of your experiment, I would recommend using limma so that you can use its duplicateCorrelation method to take into account the repeated measures.

For the design matrix, I'd combine condition and population first into condPop:

condPop patient
B1 p1
B1 p2
B1 p3
...and so on


Then use design <- model.matrix(0 + condPop) as your design matrix.

For the contrast matrix, do:

makeContrasts(interaction = (condPopB1 - condPopA1) - (conPopB2 - condPopA2), levels = design)


This will test for the population difference in treatment response.

If you want to use DESeq2, I guess you can do the same but use this model matrix instead:

design <- model.matrix(0 + condPop+ patient)


This way, you are blocking on the patient ID.

1

We have a section of the DESeq2 vignette on how to compare across groups when individuals are nested.

https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#group-specific-condition-effects-individuals-nested-within-groups

Thank you for your instruction! I will read it thoroughly and try it.