Hello,

I have RNASeq data from two conditions (control, treatment) each with 5 different time points 0d, 3d, 7d, 14d, 21d (each with two biological replicates). I see you your answer in this thread C: DESeq2 likelihood ratio test (LRT) design - 2 genotypes, 4 time points to get the genes over the time between the condition (control vs treated) and just over the time (avg. all the time point of control and treated and get the genes over the time).

I'm looking four different set of genes...

1. genes over the time between/across the condition

2. just over the time

3. contrast of each time point between control vs treated

4. genes over the time but with in each individual condition ie. DE genes over the time in control separately and the genes over the time in treated separately

For the first question I see ~condition + time + condition:time as the design

dds <- DESeq(dds, reduced=~condition+time, test="LRT")

res <- results(dds)

second question design(dds) <- ~condition + time

dds <- DESeq(dds, reduced=~condition, test="LRT")
res <- results(dds)

How would I do the one for my fourth question? over the time of genes in each condition separately

Please explain me what is the design matrix suppose to be used and the test (LRT or Wald)?

I'm thinking of use the matrix only with one condition of the data eg. all the time points of control data  and use design(dds) <- ~ time
dds <- DESeq(dds, test="Wald")

One more question, please clarify me any difference of saying/calling DE genes *over the time and all the time point* here, what represents over the time (just DE genes at every time point)?

Here's my matrix. Thank you in advance.

Library Replicate group condition time

R76 A group1 control 0d

R85 B group1 control 0d

R77 A group2 control 3d

R86 B group2 control 3d

R79 A group3 control 7d

R88 B group3 control 7d

R81 A group4 control 14d

R90 B group4 control 14d

R83 A group5 control 21d

R92 B group5 control 21d

R76T A group1 treated 0d

R85T B group1 treated 0d

R78 A group6 treated 3d

R87 B group6 treated 3d

R80 A group7 treated 7d

R89 B group7 treated 7d

R82 A group8 treated 14d

R91 B group8 treated 14d

R84 A group9 treated 21d

R93 B group9 treated 21d

My first take is that I'd stick to one model, rather than subjecting the data to so many different models. But the final analysis decision is up to you.

You can see from the RNA-seq workflow how I show to setup the LRT and also test each time point:

https://bioconductor.org/packages/release/workflows/vignettes/rnaseqGene/inst/doc/rnaseqGene.html#time

For your other questions: If you wanted to test, within a single condition, whether there are differences over time, perhaps the easiest approach would be to subset the dds to just the single condition, and then run DESeq() with test="LRT", and with a full design of ~time and a reduced design of ~1.