PWF from GOseq to adjust q-value in Cuffdiff

Question

[trakhtenberg]

Hello,

In accounting for gene length bias using GOseq, the PWF is used to adjust the p-value for determining the over-represented GO category.

Has anyone made a script which would use this PWF for adjusting the q or p value in the Cuffdiff output files (I use cufflinks) of all DEGs?

Then it would be straightforward to just select all DEGs below the updated significance value threshold for all sorts of downstream analysis (including GO enrichment, KEGG pathways, or just for having a more accurate list of the DEGs to select candidates for screening)?

Thank you.

Answer 1

[Devon Ryan]

GOseq doesn't use the PWF to adjust the p-values. Rather, it uses the PWF to construct the background distribution, which is then used in the statistical test. To use a similar method with cuffdiff you'd have to completely rewrite it, this isn't a simple correction. I won't go into the fact that what GOseq is trying to compensate for isn't an issue at the level of analysis for which one uses cuffdiff.

What I think you're concerned about is GC/length bias in library construction and variance in that between samples/experimental groups. If that's something you're concerned about, then absolutely do not use cuffdiff. There's already a package to help compensate for these sorts of things, called CQN. You can't use that with cuffdiff, but you can with edgeR/DESeq2/limma.