I find this to be a hard and interesting question. When you perform enrichment, you check if your gene list is enriched in specific gene sets.
For starters, these gene sets and their meaning are very varied depending on the database. They can be collections of loosely "related" genes in which the relationship between them is not too clear (such as Gene Ontology biological processes), they can be pathways in which there is a more defined relationship of which genes inhibit or activate the different steps (such as KEGG pathways), etc. Thus, getting an enrichment in a Gene Ontology set is not the same as getting an enrichment in a KEGG set, and so on. It would depend on the specific set. For example, for a particular pathway the down-regulation of a gene which is inhibitory may have the same effect as the up-regulation of an activating gene. Globally, gene set enrichment gives us results which should be broadly interpreted as "these are the pathways affected", and not much else. For more causal explanations about mechanisms and so on we should turn to deeper analyses of particular pathways, and experiments.
Nonetheless, there are in my opinion a couple of things one can do to "refine" a bit the enrichment results:
- Separate you input by up- and down-regulated genes, so that from your results you may also know that "things in this pathway tend to be changed in this direction".
- Use directional databases: there are some databases of gene sets whose sets are directional. For example, some of the ones in the Molecular Signature Database. Thus, one may have pathways such as "UV_RESPONSE_UP" and "UV_RESPONSE_DN" (in the Hallmark C1 database), describing genes which go UP or DOWN in response to UV light. If you find that your UP-regulated genes are enriched in the "UV_RESPONSE_UP" pathway, this may point towards the notion that, in your biological process being studied, the UV response is possibly being activated.
Both these things can be accomplished by using GSEA, because it takes into account the direction of change, and also because it finds enrichments using the MSigDB databases, some of which are directional.
modified 4 months ago
4 months ago by
Papyrus • 550