I used DEseq to find differentially expressed genes between two samples, for each sample i have 3 replicates, for the DEseq result, i got exactly same padj value for a lot of genes but their p-values are different as attached, is this normal ?

Deseq same padj values for a lot of genes but with different p-values

2

Entering edit mode

18 months ago

yanyanwu
▴
20

I used DEseq to find differentially expressed genes between two samples, for each sample i have 3 replicates, for the DEseq result, i got exactly same padj value for a lot of genes but their p-values are different as attached, is this normal ?

4

Entering edit mode

18 months ago

jared.andrews07
★
14k

To expand on German's answer, this is due to how adjust p-values are calculated with the default (Benjamini-Hochberg) adjustment procedure in R. In short, the p-values are ranked from smallest to largest, and those ranks become part of the calculation.

This Stats SE post explains it nicely, but in short, each unadjusted p-value is multiplied by the number of tests and then divided by its rank order. When p-values are particularly close to each other, this can lead to a more lowly ranked unadjusted p-value ending up with a smaller adjusted p-value than the one before it. In these cases where the resulting sequence is non-decreasing, the preceding p-value is changed to the subsequent one such that they are the same. This is what you're observing in your results, and it's normal.

3

Entering edit mode

18 months ago

German.M.Demidov
★
2.8k

Yup, this is fine. FDR shows the expected proportion of false discoveries at this threshold and sometimes for different p-values the expected FP proportion is the same for different p-values.

Similar Posts

Loading Similar Posts

Traffic: 1899 users visited in the last hour

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

if you want the really short, terribly oversimplified way I think of this, it's, "FDR is rank based, which introduces quantization"

In this case, should I still use the adjusted p-values, or I need to switch to the p-values for the filtering instead?

The adjusted p-values are still perfectly valid, and I would not recommend filtering on un-adjusted p-values for DE analyses.