DeSeq2 for time point analysis? And what is the best way to handle technical replicates

Question

[halo22]

I have the following two questions, I would highly appreciate your input on both of these.

1) How to create a design matrix for a time point comparison where the patients were on and off treatment using DeSeq2? The patients were initially on drugA for a month and then off drugA for another month, the RNA was collected before the drug intake and a month after stoping the drug.

My phenotype file for the analysis looks this:

SampleName     SampleID   Condition  

P1T1                  Patient1       DrugA

P2T1                  Patient1       DrugA

P3T1                  Patient2       DrugA

P4T1                  Patient2       DrugA

P1T2                  Patient1       Off-A

P2T2                  Patient1       Off-A

P3T2                  Patient2       Off-A

P4T2                  Patient2       Off-A

Here: P1T1 stands for Patient1 and time1(on DrugA) and P1T2 Patient1 and time2(Off-A). I have more patients but for this example I am just showing two patients with two replicates and two time points. (Here PT are columns of my expression matrix).

Can you please suggest a DeSeq2 design matrix for this?

2) Since the same patients were sequenced twice for both time points as technical replicates, does it make sense to combine the two replicates into one or consider them as two separate samples?

Please let me know if the post requires more explanation?

Thanks

Answer 1

[Devon Ryan]

1. The design is ~SampleID + Condition.
2. Yes, you will need to sum the replicates. In DESeq2 there's a collapseReplicates() function to make this easier, though you'd need to add a column to the colData like dds$grp = paste(dds$SampleID, Condition, sep="_") and use grp in collapseReplicates().