Hi, 

I am wondering how the result of GSEA using cluster profiler is different from that of GSEA GUI java application. I initially used GSEA GUI desktop application and tried clusterprofileR package in R using gseKEGG function. I assigned latest kegg database available online and pvalue cutoff of 0.05 for cluster profileR. I also assigned the same permutation number and minimum geneset size to be using the same condition as what I used for GSEA GUI software. However, the results are different; some gene sets I got as downregulated in one were upregulated in the other and vice versa. Anyone has an explanation for this and which is preferred for general practice?

Thank you for your input in advance!

clusterProfiler/fgsea differs from Broad GSEA in the multiple-hypotheses correction procedures. The latter uses an ad-hoc procedure while the former uses any standard method with Benjamini-Hochberg being the default one. I would recommend using at least 10000 permuations for  clusterProfiler/fgsea (which is still pretty fast).

In details this is covered in the fgsea manuscript (https://www.biorxiv.org/content/early/2016/06/20/060012), comments on biorXiv, and also here: https://bioinformatics.stackexchange.com/questions/149/are-fgsea-and-broad-institute-gsea-equivalent