In order to improve an existing method to test enrichment of some terms (biology), I developed a new approach using special permutations while the standard approaches are using hyper-geometric test.
The number of permutations is limited to $20000,$ and there are thousands of hypothesis. Once correcting the calculated permutations $p$-values to multiple hypothesis using $FDR$ we remain with $0$ enrichments.
In my opinion it is partially due to the different range $p$-values of our approach (max $p$-value is $10^{-5}$) and large number of hypothesis.
How can I overcome this? Is it legit to run first the standard approach, filter the terms that their $FDR$ passed the threshold, and test my new approach only on this subset? What steps should I take to make this correct?
Any other suggestions?
Thank you!