recommend gene ranking methods when doing GSEA

Question

[luckyday1661]

Dear guys,

Do you recommend to use t to rank gene or log2FC or sign(log2FC) * -log10(Pvalue) to rank gene?

The fgsea use t to rank the gene. fgsea rank gene Rscript

While when looking the distribution of the DEGs, t is similar to log2FC, but different from the others like Pvalue, adj.Pvalue, etc.

Aslo, when looking at the results from GSEA, it seems Pvalue is more mentationed, while not using adjusted-P value?

Answer 1

[rpolicastro]

I tend to use shrunken log fold changes from e.g. DESeq2 to reduce the prevalence of genes with high log fold differences but very low expression. Log fold change is also a more direct measurment of effect size compared to p-value so intuitively made more sense for GSEA.

[luckyday1661]

Thank you very much for sharing for very nice valuable experience! Appreciate it a lot!