relevance of GSEA on Nanostring ncounter gene expression data ?

Question

[guillaume.rbt]

Hi all,

I'm currently doing analysis on a gene expression dataset produced with the nanostring ncounter method, on a panel of around 700 genes (all involved in the immune system)

I want to test for gene set enrichment, based on the differential expression analysis I've performed between two conditions.

With RNAseq data I usually use the GSEA tool, would it be relevant to also use this tool with a "small" panel of 700 genes? Wouldn't it be biased, as all of those genes are all involved in the immune system?

(I think not but I'm not that sure)

Thank for any of your input !

[siRi]

Hi, I also have the same setup and would like to ask, how to deal with nanostring data and GSEA (is it statistically possible/ correct?) and whether I should rather used a preranked GSEA ? Thank you for your input!

Answer 1

[i.sudbery]

Yes, used naievely it would be biased. However, you should be able to set your "background" set to only include the 700 genes, not all genes in the genome.

To show this, I appeal to a slight simpler situation - standard hypergeometric enrichment analysis (i.e GO analysis). For example if 20% of my differential genes are in category A, but 5% of genes in the genome are, then thats a 4 fold enrichment over the genome. But if I only have 700 genes, rather than 20,000 genes that my DE genes could be chosen from, then I should use the rate of Category A in those 700 genes. Say 10% of the 700 genes are in category A, then my real enrichment is only 2 fold. However, most enrichment software allows you to set a background set to cope with this.