How to choose between doing a 'regular' GSEA analysis and a preranked one?

Question

[emibaffo]

Hello! I'm analyzing the hepatocellular carcinoma data from TCGA. I have downloaded it through the package TCGAbiolinks, then normalized it and performed a differential expression analysis using the edgeR method which is included in the TCGAbiolinks package. I now want to perform a GSEA analysis using the GSEA desktop software, in order to see which pathways and GO terms are enriched. I have two options: I can either plug the normalized data with each column being a sample and provide the phenotype file, or I could do a preranked GSEA providing the list of differentially expressed genes ranked by logFC.

I'm trying to figure out what is more convenient or more statistically accurate. I read here that preranked is ideal for when you have less than 3 samples per phenotype but it's not the case here (50 normal and 373 tumor). So is there one I should pick over the other? My instinct tells me the regular analysis is better because the preranked is biased by the previous differential expression analysis but I'm kinda just guessing on that. I'm open to any advice.

Thanks in advance!

Answer 1

[Alex Gibbs]

Performing a GSEA preranked analysis allows you to find out if you have enrichment at the top (positive logFC) or bottom (negative logFC) of your DEG list. I think in your case, where you have 373 tumour samples vs 50 normal samples, it may be beneficial to perform this analysis to give you a general overview of how your tumour dataset behaves as a whole.

On the other hand however, performing the 'standard' GSEA will allow you to observe enrichment between selected phenotypes (you may have a number of tumour samples that fall under tumour subtype A and a number that fall under subtype B for example).

It all depends on what your research question is. I personally like to do both methods (if I have time) just so that I have the data there if I need it, but I think in your case, you may benefit from performing the standard GSEA so that you can observe and potentially identify any differences or similarities between your tumours.

Hope this helps :)

Alex

[emibaffo]

Thanks! I just wanted to see which pathways were increased in tumor samples and which in normal samples but I guess that I could just do both as you said.