Question: p-value or adjusted p-value?
gravatar for R
5.2 years ago by
R10 wrote:


I used DESeq package to analyze my RNA-seq data but I have a problem regarding filtering. I have only one replicate for each sample (therefore I used method="blind", sharingMode="fit-only"))

The problem is that all padj values are 1, there are few cases which are below 0.1.  Due to the fact that I do not have replicate, should I consider padj or I can only apply a cutoff (e.g. <0.05) for pvalue?

It would be great if you can guide me


rna-seq • 6.0k views
ADD COMMENTlink modified 5.2 years ago by seidel6.9k • written 5.2 years ago by R10

As a sidenote, here is a primer that goes over multiple testing correction (what it means to adjust your p-value)

ADD REPLYlink written 5.2 years ago by Ying W4.0k
gravatar for seidel
5.2 years ago by
United States
seidel6.9k wrote:

You don't have a problem with filtering, you have a problem with your data in that you have only one replicate and thus no real measure of variation. Whatever p-values you have just use them as a ranking, and examine from the top down for what makes the most sense. It's not that uncommon for an experiment to yield results with high adjusted p-values. The ranking given by regular or adjusted p-values should be the same (for those that aren't 1).

ADD COMMENTlink written 5.2 years ago by seidel6.9k

in such cases, how much would it help to set a higher fold change?

ADD REPLYlink written 5.2 years ago by R10

If you have too many genes to choose from, taking a top set by p-value, and then ranking those by ratio may help you, but in my experience it depends on the shape of your data. You can use an MA plot to determine how the ratios look for genes with low p-values. It's not uncommon to find many genes with high ratios, and low p-values, way down at the low-signal end of your data (thus based on differences between small numbers). On the other hand, if the things with low p-values are distributed across your signal range, you may by able to use ratios. I often make MA plots and drill into the data diagonally, ranking a significant gene set by both signal and ratio. You might also find Volcano plots useful for examining the properties of your data.

ADD REPLYlink written 5.2 years ago by seidel6.9k
gravatar for Zhilong Jia
5.2 years ago by
Zhilong Jia1.5k
Zhilong Jia1.5k wrote:

Yes, you can. The cut-off of the p-value or adjusted p-value depends on you. If you need more DEG, setting cut-off larger, but resulting in more false negative.

ADD COMMENTlink written 5.2 years ago by Zhilong Jia1.5k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1548 users visited in the last hour