Interpreting the FDR q value

1,518 views
Skip to first unread message

Jenny Hadjiyerou

unread,
Aug 17, 2023, 12:47:22 PM8/17/23
to gsea-help
Hi all,

I am very new to GSEA and have opted to use the desktop app following some failed attempts at R. Everything is running smoothly, except, I am confused about how to interpret my results. Looking at the top 30 gene sets, the ones that I am interested in (I'm using reactome) are significant (looking at the nominal p-value), but the FDR q-value is about 0.08. I have three samples per phenotype and run the gene_set permutation type. Should I investigate these pathways further, or are they not significant? I guess I'm confused, as GSEA assumes FDR < 25% is significant? Any help is appreciated!

Jenny

Anthony Castanza

unread,
Aug 17, 2023, 6:49:27 PM8/17/23
to gsea-help
Hi Jenny,

GSEA recommends a FDR threshold of 0.25 when running in the Phenotype permutation mode. Generally with the gene set mode it is better to use a threshold of 0.05 just based on the nature of the test being performed. That said, a marginal result like this is still potentially worth investigating by other methods. Can I ask, what part or parts of MSigDB did you run? If you ran an extremely large number of sets, particularly if you combined multiple collections this can definitely adversely affect the FDR calculation due to how GSEA computes statistical significance.

-Anthony

Anthony S. Castanza, PhD
Curator, Molecular Signatures Database
Mesirov Lab, Department of Medicine
University of California, San Diego

Dharam Bir kashyap

unread,
Aug 17, 2023, 8:58:15 PM8/17/23
to gsea...@googlegroups.com
Hi, 

I am facing the problem with file format required to upload in GSEA. I am trying to upload a *tex file with all the changes as mentioned in the guideline, but it is not accepting the file.  I am using a Windows PC.  

--
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/74455ef8-16cb-45f3-9b23-63e202e7eb88n%40googlegroups.com.


--

Dharam Bir, PhD

Post Doc

Brown Center for Immunotherapy

Melvin and Bren Simon Comprehensive Cancer Center

School of Medicine, Indiana University

975 W. Walnut St. IB557

Indianapolis, IN 46202, USA

Email: db...@iu.edu & make...@gmail.com

Mobile: 317-384-2878

 

Jenny Hadjiyerou

unread,
Aug 18, 2023, 4:21:48 AM8/18/23
to gsea-help
Hi Anthony,

Thank you for your help, that makes sense! I guess I was just confused, because when I originally run this in R using the fgsea package, all these pathways came up as significant, with a padj less than 0.05 and I'm worried about doing something wrong. Here are my settings, and it would be helpful if someone could let me know if they are correct?

Jenny
GSEA settings.png
GSEA settings 2.png

Castanza, Anthony

unread,
Aug 18, 2023, 12:49:45 PM8/18/23
to gsea...@googlegroups.com

Hi Jenny,

fgsea uses a different statistical model for computing significance statistics, one that gives much more optimistic pValues than GSEA. Our lead mathematician has some significant disagreements with their approach but a 0.08 FDR, and a significant nom pValue, combined with a significant result from fgsea is reasonably compelling to me.

 

-Anthony

 

Anthony S. Castanza, PhD

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.

Castanza, Anthony

unread,
Aug 18, 2023, 12:52:29 PM8/18/23
to gsea...@googlegroups.com

Hi Dharam,

I’d ask that you make a new thread to discuss your specific issue.
That said, if you followed the file format guidelines, make sure that it was properly saved as tab delimited text, and then had the correct file extension (.txt, OR, .gct depending on what format you chose).

If you still encounter issues after double checking the file format, encoding, and extension, please let me know (again, a new thread please).

 

Thanks

 

-Anthony

 

Anthony S. Castanza, PhD

Department of Medicine

University of California, San Diego

 

--
You received this message because you are subscribed to a topic in the Google Groups "gsea-help" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/gsea-help/0RIVtr7aESQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to gsea-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/CAA1Kh%3DB%3DDjCZ8TeyBWjYmMXvXM16FJwTJqAE-yjsLCDfuaPoOw%40mail.gmail.com.

 

Anthony Castanza

unread,
Aug 18, 2023, 12:58:20 PM8/18/23
to gsea-help
Hi Jenny,

Sorry for the multiple emails, I missed the screenshot you attached.
I might suggest using the "Collapse" setting for Collapse/Remap, with an appropriate chip file for your gene identifiers, this makes sure that all the genes in your dataset are matched to their appropriate gene symbol version in MSigDB. If you're already using gene symbols in your data there is a "Symbol Remapping" chip file that can be used for this. That might help improve scores some. Additionally, as a note, when rerunning GSEA, even with identical settings, you might notice some variance in the significance statistics, this is due to the random seed that is used to generate the null distributions and is normal. However, if you need to reproduce a run exactly, the random seed for any given run is printed in the "Comments" section of the report index and this value can be reused in Seed for Permutation field in the Advanced Fields section of the GSEA Software.


-Anthony

Anthony S. Castanza, PhD
Curator, Molecular Signatures Database
Mesirov Lab, Department of Medicine
University of California, San Diego

Jenny Hadjiyerou

unread,
Aug 18, 2023, 1:52:39 PM8/18/23
to gsea-help
Thank you Anthony, you've been extremely helpful!
Reply all
Reply to author
Forward
0 new messages