search

MaayanLab / blitzgsea

Fast Gene Set Enrichment Analysis (GSEA) implementation of the prerank algorithm. Use Loess interpolation of bimodal ES distribution for accurate p-value estimation.
Apache License 2.0
40 stars 8 forks source link

Normalized enrichment score = negative infinity and results not reproducible across different computers #36

Open robinpaul85 opened 1 month ago

robinpaul85 commented 1 month ago

Many times I see the normalized enrichment score to be infinity or negative infinity. Also in general, i find the results to be not reproducible when I run the same input in different computers. I see the code base makes extensive use of random numbers. Maybe set a seed for all random numbers?

Screenshot 2024-10-04 at 1 35 46 PM
lachmann12 commented 1 month ago

Hi, thank you for the feedback. Can you confirm that this is happening on the latest version of blitzgsea? The latest version is 1.3.47 which did address the inf issue. If this is still something you experience is it possible to share the signature that causes this so I can try to reproduce the error?

robinpaul85 commented 2 weeks ago

Is there a way to check the blitzgsea version ?

lachmann12 commented 2 weeks ago

You can use the following command line call:

pip3 freeze | grep blitzgsea

lachmann12 commented 2 weeks ago

if you are using windows:

pip3 freeze | findstr blitzgsea

robinpaul85 commented 4 days ago

Yes, I did upgrade the blitzgsea version. It still shows NES=-inf. The most difficult part is its hard to reproduce the results even on the same computer. Sometimes it shows a non-infinite number and sometimes infinity.