Differential analysis of functional enrichments

Question

[russhh]

Hi,

Experimental setup: I've got disease vs non-disease expression data from human, and Model vs Control expression data from a mouse phenocopy of the disease.

What I'd like to do is some formal comparison of the functional enrichments (ie, GSEA, GO enrichment-type approaches) in the human versus the mouse data. I'm not aware of any such method and would like to do something a bit more robust than running two separate species-specific GO enrichment tests and making a big ugly Venn diagram.

Do you have any suggestions?

[h.mon]

Maybe rank-rank hypergeometric overlap?

[russhh]

Thanks, that's really interesting. I'm quite interested to find GO terms that are specifically enriched in one species but not the other, which seems to be opposite to the aims of the method you mention.

[h.mon]

You may try GSVA + limma:

1) use GSVA to transform gene expression data into gene-set score expression data 2) use limma to fit a linear model and test for differential expression for the desired contrast

I can't suggest details about how to exactly implement, as I never did such an analysis. For example, should you use GSVA to transform human + mouse data all at once, or should you transform them separately and use some limma function to normalize between them?