0
$\begingroup$

I am using fgsea in r to calculate and plot a bunch of GSEA graphs.

My question is whether it is acceptable to use the Wald statistic from DeSeq2 to rank the gene list?

I have seen in the GSEA application that the signal to noise ratio is used but is the Wald statistic sufficient?

If not then is there an easyish way to calculate the signal to noise ratio in R given a normalised count matrix?

Thanks in advance.

$\endgroup$
2
$\begingroup$

I personally rank by signed nominal p-value. The fgsea author recommended on biostars.org to use either -log10(nominal p-value) or the F-statistics column (or whatever statistic the tool you use outputs) but not FDR/adjusted p-values as the latter produces a lot of ties for genes with low significances.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Please note that I edited the post. It must of course be -log10(nominal P), no nominal P itself, sorry. $\endgroup$ – ATpoint Jun 3 at 15:41
1
$\begingroup$

If using DESeq2 with GSEA, I'd recommend ranking by shrunk log2FC values.

It'd also be worth considering ranking positive and negative associations separately, because the standard GSEA algorithm doesn't cross at the zero point when associations change from positive to negative.

You shouldn't be using p-values to rank anything. A p-value gives information about the reliability of a result, but not its importance. See point #5 of this article about p-values:

https://amstat.tandfonline.com/doi/full/10.1080/00031305.2016.1154108#_i31

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.