runGSEA actually perform simple overrepresentation analysis, not GSEA

[liuzhenyu-yyy]

Current runGSEA is hardcoded to perform enrichment analysis with fgsea::fora, which utilize hypergeometric test to evaluate overrepresentation of reference gene set (msigdb gene sets) within given query gene set (MP genes), see https://rdrr.io/bioc/fgsea/man/fora.html.

GeneNMF/R/main.R

Lines 566 to 571 in be6cb8f

	fgRes <- fgsea::fora(pathways = DB_list,
	genes = genes,
	universe = universe)
	fgRes <- fgRes[fgRes$pval <= pval.thr,]
	return(fgRes)

This is generally considered as GO enrichment analysis, which returns fold enrichment and P values, not GSEA analysis, which returns NES and p values. Current function and manuals might be misleading.

[mass-a]

Hi @liuzhenyu-yyy
thanks for pointing this out.

You are right, fgsea::fora() performs ORA rather than GSEA. These are related types of enrichment analyses but they make different assumptions on the data. We will update the manual for the runGSEA() wrapper and probably rename the function in a future update.

[LongpanICR]

Hi,
I am also interested in MP annotation. Here you choose ORA analysis to do the annotation, does it suggest that using gene weight/loading as rank in true GSEA analysis is not so good for MP annotation?

thank you!
best,
Long