Alternate DESeq2 designs to model donor variance

Question

[swbarnes2]

I have a set of RNASeq data; 3 treated replicates, 3 control replicates, from each of 4 donors. I observe in the PCA plot that my Treatment corresponds to PC3; PC1 and PC2 correspond to the donors. Usually, when running DESeq2 on this data, I'd use a formula of ~ Donor + Treatment, but someone suggested that instead of using Donor in the design, I could use ~ PC1 + PC2 + Treatment

Would this be valid? Would it be better than using Donor? Or just different?

[Kevin Blighe]

~ Donor + Treatment and ~ PC1 + PC2 + Treatment produce similar results? It would depend on how much variation is captured by PC1 and PC2, of course.

[swbarnes2]

Top genes were comparable. I think adjusted p values were better with PC1 and PC2 than using Donor.