Getting precise FDR values and plotting if values are 0

[ThomasThaewel]

Thank you so much for the great implementation, I just have a couple of questions here.
If I use the standard gp.gsea function on my data (merged single cell dataset), I do get a lot of terms where FDR, NOM p-val and FWER are 0.
I use permutation_num=1000 like suggested in the snRNA-seq example, I read in the closed issues that this seems to be an issue if used with ssGSEA but is this the same case here for multiple samples?
The problem I run into then is that I can't use the suggested dot plot (gp.plot.dotplot(df_fin, column='FDR q-val', top_term=15, figsize=(4,8), cutoff=0.25)) since the FDR values of all top terms are 0.
Is there a way to get more precise FDR values or enable plotting without manually changing the values?
Thanks!

[zqfang]

see this FAQ: why FDR is 0.

when nomial-pvalue is 0, as does FDR.

• one dirty trick to get a more precise pval/fdr is that replace the 0 with (1 / num_permutations).
• or increase the permutation number

P.S. for prerank analysis, you may refer to use blitzgsea, which use an approximation method to fit a gamma distribution and you'll get very small pvalue.