Open Tomnl opened 2 years ago
GSEA use permutation (gene label or class label) to generate background distribution and calculate p values.
From my perspective, I don't recommend you to combine differenct background distribution and perform BH correction ( just like you did (concat res1 and res2 and do BH correction), given rnk1 and rnk2 is not the same.
I have never try to compare the null distritbution using different rankings. if the null distribution are all similar, I think BH correction works perfect.
I was wondering if there was a recommended way of using GSEApy when the analysis involves comparing more than 2 sample classes. e.g. control Vs class1, control Vs class2 ..... etc
It would be quite straightforward to perform repeated analysis using either prerank or with the gsea function. But I believe the FDR result might not be appropriate...
Potentially I could just combine the resulting dataframes and recalculate the FDR value using the Benjamini-Hochberg approach - see below. But I just wanted to check if this is appropriate and/or if it could be done another way within GSEApy, potentially using the gsea_fdr function.
Setup
I am using GSEApy version 0.9.16, Python version 3.9.7, and operating system Ubuntu.