Off topic:DESeq2 model design + contrasts (not full rank)

Question

[caggtaagtat]

Hi,

I have a question to the DESeq2 contrast. I have single end reads from 16 samples with 4 treatments (group) and 4 biological replicates (indi). However, the replicates are not evenly distributed across the used flowcells (flow). Can I still correct for the flowcell batch effect?

My meta table looks like this:

group indi flow 
  T1   I1   A     
  T1   I2   A     
  T1   I3   B     
  T1   I4   B     
  T2   I1   A     
  T2   I2   A    
  T2   I3   B     
  T2   I4   B     
  T3   I1   A     
  T3   I2   A     
  T3   I3   B     
  T3   I4   B     
  T4   I1   A     
  T4   I2   A     
  T4   I3   B     
  T4   I4   B

I followed the instructions for such case from the DESeq2 manual:

ds_txi <- DESeqDataSetFromTximport(txi = txi_salmon,
                                   colData = meta,
                                   design = ~ indi+group)

ds_txi$indi_n <- c("I1","I2","I1","I2","I1","I2","I1","I2","I1","I2","I1","I2","I1","I2","I1","I2")

meta$indi_n <- c("I1","I2","I1","I2","I1","I2","I1","I2","I1","I2","I1","I2","I1","I2","I1","I2")

meta$indi_n <- as.factor(meta$indi_n)
ds_txi$indi_n <- as.factor(ds_txi$indi_n)

ds_txi <- DESeqDataSetFromTximport(txi = txi_salmon,
                                   colData = meta,
                                   design = ~ flow + flow:indi_n + flow:group)

Resulting in following meta table:

group indi flow indi_n
  T1   I1    A    I1
  T1   I2    A    I2
  T1   I3    B    I1
  T1   I4    B    I2
  T2   I1    A    I1
  T2   I2    A    I2
  T2   I3    B    I1
  T2   I4    B    I2
  T3   I1    A    I1
  T3   I2    A    I2
  T3   I3    B    I1
  T3   I4    B    I2
  T4   I1    A    I1
  T4   I2    A    I2
  T4   I3    B    I1
  T4   I4    B    I2

With following contrast, I get the difference between treatment T1 and T2 within Batch A:

dds<- DESeq(ds_txi)
res<- results(dds,contrast=list("flowA.groupT1","flowA.groupT2"), alpha= p_adjust_treshold,  lfcThreshold = L2FC_treshold)

But how can I get the general differences between treatment (group) T1 and T2 with elimination of the batch effect, if thats possible?

Could I maybe just do something like this:

res<- results(dds,contrast=list(c("flowA.groupT1","flowB.groupT1"),c("flowA.groupT2","flowB.groupT2")), alpha= p_adjust_treshold,  lfcThreshold = L2FC_treshold)

PCA plots: