How do you compare GSEA results between treatments?

Question

[shawn.w.foley]

I'm interested in comparing the genes and genes sets regulated by two different drugs. I performed a microarray on Drug A vs control, and Drug B vs control. Using Limma, I calculated the differential expression for all 20,000 genes, and performed a pre-ranked GSEA. I now have ~700 gene sets that are depleted by Drug A, and ~800 gene sets that are depleted by Drug B (FDR<0.05). There are ~280 gene sets that are depleted by both Drug A and Drug B.

Taking the pre-ranked GSEA output, how can I meaningfully compare these 280 gene sets depleted by both treatments to characterize similarities and differences between these treatments.

If I am interested in PI3K signaling, could I compare the Normalized Enrichment Scores (NES) for all of the gene sets involved in PI3K signaling and perform a Wilcoxon test to determine if there is different enrichment between the two treatments?

Can I take the number of genes within a gene set regulated by each drug and meaningfully compare those gene sets?

Thank you for the help, a few pointers or a publication will set me on the right track. I've searched and can't find analyses like this.

Answer 1

[Kevin Blighe]

I believe that gene set variation analysis (GSVA) will work here. This will allow you to compare two sets of enriched terms and pathways. It may take a day to learn the methods, but the tutorial is a good starting point.

Other than that, just a manual 'human' comparison by looking over the top enriched terms should suffice. At the end of the day, we have to learn to disengage the computer and start to interpret results ourselves.

Answer 2

[a_liberzon]

Why not trying two runs of standard (not pre-ranked) GSEA? Prepare a single GCT file with all your data, such that rows are probe set IDs, columns are samples for all perturbations (treatment A, treatment B and Controls). Next define a CLS file with the relevant 3 phenotype classes. Finally, run GSEA on each pairwise comparison, e.g. A vs Controls, then B vs Controls. Use the same number for permutation seed and you should be able to compare NES scores.

Answer 3

[a_liberzon]

Another thought along the same lines. Once you have GSEA results as described above, you can compare ES (not NES) scores say, by a paired tTest for each of the enriched sets.