Multiple Conditions Testing in DESeq2 with unequal number of replicates

Question

[realnewbie]

Hi, I read previous posts about multiple conditions testing. However, my case is a little bit different. I have 3 different groups (plus their controls) with unequal number of technical replicates, so I have trouble to how to compare them.

First I want to compare each treated sample with its control, then normalize them to each other.

Then compare each treated group with each other. (Do I really need that intra-normalization to compare each treated group)

Samples are like that:

treated1 control1

treated2.1 treated2.2 control2

treated3.1 treated3.2 control3

I think the design below is not correct:

                set     condition
treated1     treated1   treated
control1     control1  control
treated2.1  treated2   treated
treated3.1  treated3   treated
treated3.2  treated3   treated
control2     control2   control
control3     control3  control
treated3.2  treated3   treated

Can you help me to fix this. Thanks

[russhh]

could you explain why you needed to generate three techreps for your treated-3 sample. Are they from different sequencing batches or different lanes or something. The standard way to combine techreps is just to sum them all up

[realnewbie]

same cell lines with same treatment, but from different wells.

[russhh]

So really your experiment is designed like this:

               set     condition    cellline
treated1     treated1   treated    1
control1     control1  control    1
treated2.1  treated2   treated    2
treated3.1  treated3   treated    3
treated3.2  treated3   treated    3
control2     control2   control    2
control3     control3  control    3
treated3.2  treated3   treated    3

Can't you just model it using ~ cell_line + treatment? Or, even better, use duplicateCorrelation inside voom

[realnewbie]

Yes,you are right. Thanks for your time. However, I was wrong. Sorry for misleading to you. Each treatment is different from each other (but the .1 or.2 meaning relicas for same treatment conditions from different wells),so each control of each treatment is also different(i.e: targeting vector vs empty vector, but with different vectors).. treated1 to control1, treated2s to control2,and treated3s to control3 will be compared. It is more like ANOVA. So, I though, edgeR's GLM approach would be better for this comparison.