How does DEseq calculate p value when n=1?

Question

[RNAseqQ]

(Setting aside how useful n=1 DESeq is, this question is purely about the algorithm)

I know the p-value is related to the fold change of the two samples which I am comparing, but since there are no replicates, what else determines the p-value? In my data there are genes which have exactly the same fold change, but different p-values and all I can see is that these genes have a different estimated expression (lower expr = higher p val), so does the DESeq algorithm take this into account?

[Dave Carlson]

I realize that this is not a direct answer to your question, but you should not expect to have any reliable stats from DESeq if you don't have biological replicates. This has been discussed extensively. See for example: How to run the Deseq2 tool without replicates.

Also the latest DESeq2 vignette gives a rather clear answer regarding whether or not replicates are necessary:

Can I use DESeq2 to analyze a dataset without replicates? No. This analysis is not possible in DESeq2.

[Carlo Yague]

For clarity, actually DESeq2 allowed to analyze datasets with only one replicate. This feature was completely removed in DESeq2 0.1.22, hence the discrepancy between older posts and the current DESeq2 vignette.

Answer 1

[ATpoint]

Yes, the magnitude of the counts (=typically the expression level) is a notable factor that determines the significance. Read about MA-plots, you will see that significance typically trends to be lower the more you go to the left on the MA-plots corresponding to lower average expression (or counts in general).