GSEA on preranked list with weighted enrichment statistic (exponent=1) in R
1
0
Entering edit mode
5.7 years ago
cmgraef ▴ 10

Hey there!

I'm currently working on a functional analysis of RNAseq data. I've used the GSEA tool by the Broad Institute to obtain quick results and now I'm trying to reproduce the results with R for visualization purposes.

The fgsea is amazingly fast but apparently there is no way to calculate a weighted enrichment statistic which is why my results differ from the original analysis with the GSEA tool. Then I've looked at the DOSE tool as this allows for a weighted enrichment statistic. However, here I can't use the MSigDatabase gene set collection. ClusterProfiler doesn't seem to provide a GSEA preranked option.

Did I miss a function of these packages that would allow me for a GSEA on a preranked list with the MSigDatabase gene set collection and at the same time gives me the option to calculate a weighted enrichment statistic? Is there any other way to solve this problem?

Thank you very much for help in advance!!

Cheers, Moritz

fgsea DOSE RNA-Seq • 2.9k views
ADD COMMENT
0
Entering edit mode

Any ideas? Thank you so much!

ADD REPLY
0
Entering edit mode
5.7 years ago
alserg ▴ 1000

Could you clarify what exactly you can't reproduce? Both DOSE and ClusterProfiler use fgsea as a backend, so if it's possible with DOSE it should work fine for fgsea.

ADD COMMENT
0
Entering edit mode

alserg sorry for hijacking this thread with an unrelated question but I do not think this one here merits a separate thread. What is your opinion on FDR correction? Standard GSEA uses FDR of 0.25 as cutoff as explained in their manual. Can you comment if this is also appropriate for fgsea or if the more standard 0.1 or 0.05 should be chosen.

ADD REPLY
1
Entering edit mode

Actually, that's an important question. Broad GSEA recommendation for 0.25 is due to their ad-hoc correction procedure they developed because they didn't have enough accuracy for standard correction procedures. In fgsea we use standard Benjamini-Hochberg, so standard thresholds can be used. I even recommend to go for pathways with padj<0.01, as GSEA is known to be oversensitive.

ADD REPLY

Login before adding your answer.

Traffic: 2136 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6