Proper design in deseq2

Question

[tianshenbio]

I have 32 samples with the following conditions:

16 dry season (DS) forms: 4 stages (A, B, C, D) x 4 replicates 
16 wet season (WS) forms: 4 stages (A, B, C, D) x 4 replicates

I have two questions: 1. How does 'seasonal form' affect gene expression 2. How does 'stage' affect gene expression.

For Q1, I simply use all 16 DS samples and all 16 WS samples to find DE genes between the two 'forms':

design = ~ Form

For Q2, I take pair-wise comparisons between stages. For example, first I examine DE genes between A and B, I use 8 A samples (4DS+4WS) and 8 B samples (4DS+4WS) and only consider 'stage'.

design = ~Stage

Then I repeat this for B vs C, and C vs D. Then I take the union of the DE genes from the three pair-wise comparisons as DE genes.

Is this a proper design?

Answer 1

[swbarnes2]

I think you'd be better off with design ~ Form + Stage. (Order does not matter if you specify the contrast when calling results)

Omitting Stage from the design means that the program takes all your SD samples as replicates of the same thing with a lot of variation. If you include Stage in the design as well, the software will understand that some of the variation between samples comes from the differing Stage, and it will adjust the model to account for that, and you will get slightly better results.

Then I repeat this for B vs C, and C vs D. Then I take the union of the DE genes from the three pair-wise comparisons as DE genes.

Look up how to run an LRT test in DESeq. That might be more what you want.

[tianshenbio]

Hi, thank you for the clarification. When I use Form+Stage instead of Form alone, I got more DE genes, As you said, the program now considers differences due to different stages so I got more DE genes that were previously masked by the stage difference. And I can see most of the results overlap between the two design.