How to perform GSEAPreranked analysis in a proper manner?

Question

[Deepak Tanwar]

Hello!

I want to perform GSEA analysis on a list of genes. Out of the following two options, which one should I use and why?

1. List of all genes with their log-fold-change values.

2. filtered list of genes with their log-fold-change values. Here, the filter could be p-value, FDR or filtering based on absolute log-fold-change of 0.5

To me, option 1 might give some false positives. And, option 2 might be biased as I am already filtering out the genes with the arbitrary cut-off.

[Prakash]

Generally people take account of both fold change and p-value together for ranking gene list. Have a look at this post

[Buffo]

What about webgestalt's papers defining pre-ranked gene lists?, but also, why do you consider arbitrary a FDR cut-off?

Answer 1

[mikhael.manurung]

Hi Deepak,

Log-fold change does not take into account the variability of the change. Therefore, t-value is preferable for ranking your genes. Or you can also use p-value multiplied with the sign of log-fold change for the ranking.

Pre-filtering your genes based on any criteria beats the purpose of GSEA. The method was designed in the first place to eliminate any cutoffs whatsoever so that subtle change in the ranking of the genes between conditions can be detected.

I hope it helps.

[igor]

Log-fold change does not take into account the variability of the change

That may not always be true. For example, the fold change calculated by DESeq2 is "shrunk" using statistical information.