Generic implementation of GSEA-P algorithm?

Question

[lillian.thistlethwaite]

I'm looking for a generic implementation of GSEA so I can easily modify it for applications outside of gene set analyses. I don't want a non-bioinformatics answer, like "go to this website, input your gene sets and then press enter".

However, the original author's implementation is 2500 lines of R code, much of which seems like fluff. I'm looking for a "stripped down" concise implementation of their GSEA.1.0.R script found at the below web address, which works with data matrices in lieu of "gene databases" like their .gmt files and indicator variable integer vectors in lieu of "phenotype databases" like their .cls files. I thought I'd ask the community before I waste 2-10 hours stripping down their code share.

Broad Institute's GSEA R package download page: http://software.broadinstitute.org/gsea/downloads.jsp

Answer 1

[Jean-Karim Heriche]

Have a look at the R package GSA (accompanying paper on arXiv). This is not exactly GSEA but is supposed to improve on it. According to the authors, the differences with GSEA are:
- use of the mean of the positive or negative part of the gene scores with largest absolute value.
- different null distribution for estimation of false discovery rate.
- can handle more than two outcomes and quantitative ones.

There are papers (such as this one) that argues that one can get the same results as GSEA in a simpler way (if one is willing to make some assumptions).

For an application outside gene sets, I would consider whether simpler approaches would be applicable.