Gene set to perform GSVA

Question

[CTLong]

Hi all,

Let's say I performed DE analysis between treated and control groups. After that, I performed an over-representation analysis on the upregulated genes and found that the a specific MsigDB hallmark was significantly over-represented. What I wanted to do next is to verify this enrichment at an individual sample level by performing GSVA. My question is, whether I should use the entire geneset of that hallmark to calculate the enrichment score, or should I use only the upregulated genes to calculate the enrichment?

Answer 1

[ATpoint]

Validation/verification should always be fully independent of the analysis you want to validate, hence you go with the entire geneset.

[CTLong]

Thanks for the reply. I gave it a try and found that if I use the entire geneset, the hallmark does not seem enriched. But when performed on a subset of this gene set, it shows very high enrichment. Therefore my interpretation is that using the entire geneset seems a bit too harsh, since the hallmark has to be very significantly altered to be detected via this approach. I guess using a subset of the geneset allows for detection of more subtle hallmark changes? Wondering what is your take on this.