Hi Everyone,

I am using DESeq2 to calculate DE genes in a dataset of 6 tests and 6 control samples. The cDNA library preparation has been performed in 2 batches. The second batch contains only 2 samples and both of them are in the control group. When I am trying to correct for the batch effect (~ batch + treatment), the results are not different than when the effect of batch is not included (~ treatment). As there are only 2 samples, both in the control group, it seems there are not enough data points to estimate the effect. Does anyone have a suggestion on how to solve this problem?

I think it might not be possible to solve this, even if you had 100 additional control samples. Imagine if the control samples of batch 2 would actually be biologically different from the control samples of batch 1, there is no way to know. If you attempt to remove the 'batch' effect here, you remove any difference the two control batches have, without knowing what is batch and what is biologically relevant. So you just artificially change the values of these new points so that they are close to those of batch 1. So as far as I know the samples from batch 2 cannot be used in this analysis at all.

My best guess is that DESeq2 sees that this type of batch effect removal is invalid and just doesn't do it. Maybe you even see a warning?

You can include those samples, they will provide some very minor contribution to the dispersion estimates (one degree of freedom). You should use the design ~batch + treatment. But because you need to use one degree of freedom for the batch difference and there are only two, these make nearly no difference to the results.