This is a test version of Biostars. For the public version, visit https://www.biostars.org.
gsea ranking genes

Hello,

I’m a bit confused and I’d appreciate your help.

I have a gene lists produced with PyDESeq2, edgeR, or limma, and I would like to perform GSEApy. For gene ranking, should I rely on the stat column, or should I use a combined metric based on log2 fold change and p-value

second question in GSEA . does FDR = padjust = qvalue ?

thank you in advance

gsea rnaseq
• 821 views • link
updated by DareDevil DareDevil • written by larifi.nour2 • larifi.nour2

2 answers

The question of the ranking has been asked many times before. There is no good argument pro or contra stat vs logFC vs these composite metrics that you mention. I usually use stat as the logFCcan be strongly biased depending on noise.

For the second question, yes FDR means multip-testing adjusted p-value.

• 0 views • link
written by ATpoint ATpoint

the logFCcan be strongly biased depending on noise.

Just to add, if log-fold change shrinkage was used, noise in NS genes ought to be minimized and log2FC would be fine to use. In principle.

• 0 views • link
written by txema.heredia • txema.heredia

It penalizes noise to zero, depending on the dataset, that can reduce almost all genes to zero, so it's not universially applicable, and largely restricted (at least by direct implementation) to DESeq2 though.

• 0 views • link
written by ATpoint ATpoint

If your DE results include a proper test statistic (like the stat column from DESeq2/limma/edgeR), it’s usually best to use that for GSEA ranking since it already reflects effect size and variance. GSEA originally uses metrics like signal-to-noise ratio or t-statistics for ranking, so the stat column fits well. Combining log2FC and p-value is more of a fallback when a test statistic isn’t available.

• 0 views • link
written by DareDevil DareDevil

Log in to answer this question.

More posts like this