I am a frequent user of DESeq2 (when I don't have technical replicates of my libraries). When I have technical replicates I usually go for Sleuth.

I have seen people that conduct analysis in DESeq2 with ONLY two either biological or technical replicates.

We always include at least three samples in each group (experimental and control), and ideally more.

Is a DESeq2 comparison of two groups, with two libraries in each group, even a valid comparison? What are the most likely statistical flows in such analysis?

I am asking that because I am reanalyzing data from published studies, but few of them have only two replicates per group.

Any comment would be highly appreciated.

hi Juan,

Yes, as long as there is one "degree of freedom" DESeq2 can estimate dispersion. This is the number of samples minus the number of coefficients, so here you have four samples and two groups (represented by two coefficients), which gives two "degrees of freedom". So DESeq2 could also run with 1 vs 2. The downside is that the statistical power (sensitivity) drops off very sharply in this range. So you are only able to find the very largest fold changes. See the Schurch 2016 paper about biological replicates in RNA-seq.