FDR equal to 0

[Stephane Rodriguez]

Dear all,

I have a question regarding FDR. I ran a ranked GSEA and I indeed obseve a clear enrichment of my gene list among a specific cell type signature. However I have a p value=0 as well as FDR and FEWR=0. I tried to increase the number of permutation up to 70000 (meaning that signifiance is below 1/70000) but it did not change anythin, all values are still=0. I've red that we can change FDR from 25% to 5%, where can we set FDR threshold and will it change anything? Any idea to get FDR or FWER value?

Many thanks for your kind help and best regards

[Anthony Castanza]

Hi Stephane,

Looking at this ranked list, it would appear that you've used some sort of artificial threshold when preparing this data for GSEA, either a log2(fc) or pValue threshold. GSEA should be run with the full dataset including all genes that were expressed and not an arbitrarily thresholded one. GSEA uses the information from the non-differentially expressed genes as well in computing enrichment scores and significance statistics. In particular the information from these genes is needed in Preranked/gene_set permutation mode as they provide a substantial amount of information to the null distribution.

I would suggest re-running your analysis with the full dataset and then checking if you're getting reasonable pValues/FDR.

In regard to your specific question "I've read that we can change FDR from 25% to 5%, where can we set FDR threshold and will it change anything?". There is no setting in GSEA for this, it is just the threshold the user should consider for "hits" in the GSEA result, which should be 0.25 in the much stricter phenotype permutation mode in GSEA, which is not available in GSEA Preranked.

-Anthony

Anthony S. Castanza, PhD

Curator, Molecular Signatures Database

Mesirov Lab, Department of Medicine

University of California, San Diego

--
You received this message because you are subscribed to the Google Groups "gsea-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gsea-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/50b7ff87-4aca-40fb-89a6-4e73306c98f5n%40googlegroups.com.

[resa BMT]

Thank you Anthony for your reply,

I redo with changing the metric of my ranked gene list using log2FC instead of FC, changing the visual impression of a threshold (while none were applied). Unfortunately it didn't change anything, I still having my p, FDR,  FWER = 0...

[Dong]

Hi Anthony, 

Sorry to chime in.  I ran many GSEA analyses and found that, most reports always have some enriched genesets that have FDR equal to 0.  How to explain that?  Is it possible that FDR value is too small to show?  

Thanks.

Dong

[Anthony Castanza]

Hello,

It is entirely possible to get a Zero pValue/FDR in GSEA, particularly for very strong enrichments. Because GSEA generates a null distribution empirically, if the permutation matrix never contains a "false" enrichment as strong as the observed enrichment, that is a result of 0/nperm=0. Sometimes increasing the number of permutations will cause a false discovery to appear, but sometimes based on the dataset properties there just will never be one.

As an aside: I would advise using the log2fc, rather than the original fold change run, this better matches the data distribution that GSEA expects.

-Anthony

Anthony S. Castanza, PhD

Curator, Molecular Signatures Database

Mesirov Lab, Department of Medicine

University of California, San Diego

[resa BMT]

I agree, it happens to me also quite often, while with prospective GSEA usage, never... I guess it's also depending on how many genes are within the list?

Thanks all for your reply,

Best regards,

Steph