Appropriate minimum gene set size for gene set enrichment analysis

Question

[Bine]

Good afternoon,

I was wondering if anyone could advice me on the following:

I am running a gene set enrichment analysis (using FGSEA & HTSanalyzer, but could apply to any tool) and I was wondering how do I choose the minimum number of elements shared by a gene set and the input total genes?

I guess it depends on size of the gene set.. So for example for Hallmark gene sets - how would you choose the minimum number? After differential expression analysis (& filtering) I have around 18.000 genes left in my input dataset.

Depending how I choose my minimum number the results vary significantly!

Thanks so much for any advice on this!!

Answer 1

[yura.grabovska]

This is very much an empirically defined parameter. It would make sense to benchmark within your data. Luckingly methods built on top of fgsea are very fast so I'm not sure it would be a very onerous task to try a few different minimal gene thresholds. You say the results vary significantly, but does that mean you typically get the same ranked genesets but the p-values/FDR results are different or do you get vastly different rankings of enrichment?

[Bine]

Thank you for your answer. The adjusted p values change, which make more or less pathways significant.