Gsea To Find Function Of Predefined Set Of Genes?

Question

[s.ritchie73]

I have identified groups of genes from whole-transcriptome expression data that are associated with a couple of phenotypes. I would now like to determine the biological function of each of these gene sets.

Is GSEA the appropriate tool for this? I've tried reading through the documentation, but it seems primarily geared towards discovering gene sets for case-control data, rather than identifying the function of pre-defined gene sets.

Answer 1

[Devon Ryan]

I would tend to prefer using co-expression or something like that. Just google/pubmed for "gene co-expression prediction" for a good number of papers.

[s.ritchie73]

I've used co-expression to identify my gene sets (using WGCNA), now I'm trying to find out what those gene sets actually do.

Answer 2

[vj]

Looking at over-represented gene ontology terms could be useful. You can try DAVID which is simple but rich in features.

Answer 3

[seidel]

GSEA could be a great tool for doing what you want, if you have enough data. If you run your gene sets through many data sets, and they show enrichment under condition X, then you can examine the experiment behind condition X, (or what other groups pop up under condition X), and infer that your genes have some function in that process. Conceptually, given groups of genes put together for any particular reason, and an infinite number of experimental conditions, GSEA should be able to hand you the set of conditions that would give rise to that group, i.e. the function of the group. But this assumes you have access to a large number of gene groups and large numbers of data sets. Jill's original implementation of GSEA didn't approach it this way (the idea of groups was much more controlled). I say, any assembly of genes for any reason is a group, and can be evaluated in any data set for "enrichment".

On the other hand, you might be able to evaluate your groups by seeing how individual genes co-cluster with other known genes across an ensemble of data, as mentioned by dpryan79. If your unknowns cluster with a set of genes with known functions, you can infer that your unknowns may participate in those processes, and since you have "groups" of genes you can perhaps evaluate how tightly your group stays together, versus having each member of the group move to individual clusters.