RNA-seq analysis design

Question

[illinois.ks]

Dear all,

I am analysing our RNA-seq data to find DEG responding our drug combinations and have a few questions.

I have 4 conditions and per each condition I have 3 replicates

group 1. Control (3 replicates)
group 2. Disease model (3 replicates)
group 3. Disease model + 1 drugs ( 3 replicates) 
group 4. Disease model + 1 drugs + 1 adjuvants  ( 3 replicates)

My hypothesis is that the drug is effective, and hopefully with adjuvants, it is much more effective....

In order to test this hypothesis I would like to compare several things.

comparison 1.  group 1 vs group 2 ( the genes affected by disease) 
comparison 2. group 2 vs group 3 (the genes affected by drug ) 
comparison 3. group 3 vs group 4 (the genes affected by adjurvants)  
comparison(maybe) 4. group 1 vs group 3 (expect that there are not so many DEG if drug is effective) 
comparison (maybe) 5. group 1 vs group 4 (expect that there are not so many DEG if drug+adjurvants is effective)

First of all, I made a dataset with these 4 groups of data to find DEGs responding one of above comparions. (by adjusting other factors)

However, it seems that it brings too many variances due to the group 2 (which is a disease model ) Note that other three groups indicate the cure model ( recovered from disease)

Could you please someone recommend that whether I just need to compare only each of two groups? ( For example, for comparison 1 , I can just use two data set from group 1 and group2 etc.. )

Answer 1

[Devon Ryan]

You're usually better off fitting all of the groups at once (a design like ~group) and using contrasts for the various comparisons, since you get the best variance fit. That is, don't subset the data for each comparison, just fit the whole thing.