Gsea Pre-Ranked File Format

Question

[jerome.lane.34]

Can I use p-values in the pre-ranked file provided to GSEA? The format of the file would look like (Gene symbol -- p-value):

HGMGCS1     1.83E-13
TMEM97      2.14E-11
..
..
NDST4       0.546

[SimonD]

The files you are providing are gene-expressions, gene-sets and a grouping arent they? Where do you want to use the pvals? Maybe instead of expression-levels?

[jerome.lane.34]

I provide only the pre-ranked file to see if gene sets database used by GSEA are enriched by my genes which are in my pre-ranked file.

Answer 1

[siabadaba]

The biggest problem of using strightforward p-value is that youy loose information about up/down regulation. Personally, I've been satisfied with the result of my analysis using this (p-value-based) way of computing rank ordered gene list:

http://genomespot.blogspot.com.au/2016/04/how-to-generate-rank-file-from-gene.html

Answer 2

[David Westergaard]

If you read the documentation you'll notice the a few sentences directly applicale to your case:

For instance, you might have used your favorite tTest-like statistic to produce a ranked ordered gene list from your dataset which you now want to test for enrichment. Order of lines does not matter. It is important, however, that the second column will have numeric values - they will be used to rank order genes by GSEA.

[jerome.lane.34]

I do not understand what they mean by tTest-like statistic. Does this includes Wilcoxon signed rank-test ? (This is the test I am using to get my p-values). I addition, if I give the gene list with p-values as describe above, gene will be ranked according to p-value in a decreasing order. This means that the most significant gene in my list (HGMGCS1) will be last.

[David Westergaard]

What they imply is that it does not matter HOW you came up with the numbers. What does matter is that the second column have numeric values, as they are used to rank the genes.

What it boils down to, is that YOU have to make a judgement whether the Wilcoxon signed rank-test is a sensible way to calculate a p-value used to rank your data.