GSEA and Cytoscape Enrichment maps

Question

[kdca]

Hi everyone,

I'm working with human RNA-seq data and performed a GSEA using the software (GSEA 4.3.3). When I try to export the results as an enrichment map to Cytoscape, it asks me to set two cutoffs: one for the p-value and one for the FDR. Since the FDR is calculated from the p-value, I'm wondering — wouldn't applying both filters be redundant? Thanks in advance for your help!

Answer 1

[Bastien Hervé]

p-value and FDR are working in pairs.

Genes pvalue from null are expected to be distributed uniformly (even with pvalue < 0.05), so the more (True Positive) your gene will be the smallest the pvalue it will have. FDR in the other end, set a fraction of comparisons for which the null is true.

From here

EnrichmentMap tips on parameter choice

Answer 2

[ruth.isserlin]

Both options are available in the Enrichment map mostly for convenience. In some cases, where signals are weak, none of the gene sets pass any reasonable FDR threshold, there is still the option to filter the results by p-value.

Also, the FDR in GSEA is not just the corrected p-value.

The p-value in GSEA is the empirical p-value calculated when comparing the ES score of a given geneset to the background distribution of ES scores from N randomizations of the data or genesets.

The ES is subsequently normalized using this randomized distribution to account for gene-set size to generate the NES (normalized enrichment score). NES = ES/(mean(abs(ES null))

The FDR, on the other hand, is the probability that a gene set with a given NES is a false positive and is calculated using the following formula - FDR(NES(obs)) = number of null NES >= NES(obs) / number of observed NES >= NES(obs)