Relate genes to a particular tissue / organelle / process

Question

[Kevin Blighe]

I thought to put this to the community:

Are there any proven ways for relating lists of genes to particular tissues, organelles, or processes? It cannot be as simply as performing enrichment against Gene Ontology or KEGG, can it? The risk of performing a basic enrichment is that you then 'cherry pick' the findings.

Another route is going the 'tissue specific' route and utilising datasets like FANTOM5 and GTEx to say that, e.g., these genes are definitively expressed in CNS or mitochondria.

Kevin

Answer 1

[jared.andrews07]

Yeah, enrichment analysis has a bad tendency to be misused pretty frequently for just the reason you describe. If you show all of the tissues/organelles/pathways, it's less of an issue, as it's much more believable if the top 10 hits or whatever are all related. But if you harp on a singular outlier (e.g. golgi bodies when most of the other hits are related to the ECM/cell membrane), it becomes much more apparent that you're "cherry picking". This is especially true given the nature of enrichment analysis where 3-5 genes can be considered a significant enrichment depending on the term and the number of genes attached to it. In general, if you're just trying to show a general association with a given process/pathway/tissue-specificity, I think just showing the top x hits from your enrichment analysis is usually enough to get the point across, even if every single term/tissue isn't exactly what you want. Beyond much more than that, you'd best start tying in some experimental validations.

As for other options, the tissue-specific route is a pretty good one, though more labor and time intensive. As you mention, GTEx has a wide variety, and other resources like the Human Tissue Atlas and JENSEN Tissue Database give more experimental protein data. The Human Cell Atlas provides protein-validated cellular localization data that's pretty handy.

[Kevin Blighe]

Thanks Jared. That is a good enough answer for me.