Hello,

I have seen examples of design formulas that have the replicate factor and others that do not use it. I was wondering when one should put the replicate factor in the design formula and when not.

For example, I have a dataset that has independent inoculations of whole plants in the greenhouse. We expected the replicates to be variable and the PCA showed it (of course they are less variable than the effect of the inoculation). So I tested with and without the replicate in the formula and without the replicate, I cannot get to see any interesting DEGs but if I put the replicate factor in the design formula, I see the genes that are expect to move.

Then, we have another data where the biological replicates are just branches so they do not vary that much (which was observed in the PCA) and in that case adding the replicate factor did not bring further information.

So what would it mean to put the replicate factor in the design ? Is it recommended to do it when the replicates are very variable ?

Thank you for you enlightenments !

It really has to do with your experimental design, and I recommend that for complex designs, people consult a statistician for choosing design and interpreting results.

If the replicates are nested within the condition of interest, then DESeq2 cannot incorporate them, due to confounding.

If they are not nested within, I would typically add the replicate/donor/cell line to the design analogous to a “paired” design.