Bioinformatics : Drawbacks of using ORA(Overlap Analysis)

Question

What do you think are the potential drawbacks/weakness of using ORA to explain distinction between two phenotypes.

I identified a few which were the dependencies of DE and the statistical method used to filter the starting DE list.

Any other weakness?

Thanks

Answer 1

It really depends on the specific methods you're using. If you're using a pathway-based methodology, I'd strongly recommend reading Khatri et al (2012) (1), which provides a good overview.

The main issue that's not touched on in that work is that pathway-based methods are highly dependent on the pathway definitions used, and the contents of the pathways are not always what you might expect in terms of their topology (a good example is KEGG's Neuroactive ligand-receptor interaction (2), which contains a huge number of events that look like 'single ligand binds to receptor', with no downstream events).

Answer 2

Like Liam said, this sort of analysis is dependent on the database you use, which may not be adequate for what you are trying to accomplish. And as far as I know, these databases are at the gene, not isoform, level. So if a change in two conditions is a result of a change in which isoform is expressed, rather than the amount of expression, you will not detect it.

Also it treats genes as equals that are independent of each other but thats usually not the case. For example if p53, at the top of a pathway, is expressed differently, that has much stronger implications than if a gene at the bottom of the pathway was expressed differently. Additionally, it treats each pathway as independent of each other, but again that is usually not that case and there usually is some overlap between pathways.

You can read more here http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1002375