Trouble understanding seemingly contradicting results of GSE analysis (contradicting enriched GO terms with similar negative NES scores)

Question

[Manko47]

Hello everyone. Since I'm struggling to find any thread related to what I want to ask I have decided to create a new post. I recently did start some simple GSE analysis to get a quick overall look at my data results before going more into details and I have trouble understanding one thing.

After running it on my DESeq2 results utilising the package ClusterProfiler and function gseGO and watching some tutorials about how to do it in R I got plenty of GO terms that are enriched in my dataset with varying ES/NES, and Padj scores. Results are mostly what I expected since at the very top immune related GO terms are found. The question that I want to ask however is that near the top I found 2 GO terms enriched - that is "Positive regulation of immune response" and "Negative regulation of immune response". Both of those have similar negative ES/NES scores. And plenty "core enriched" genes are common between those two.

Now my first thought was that both of these contradict each other. I mean the first one is telling that immune related things are rather going down (genes associated with positive response are rather down regulated), and the second that they're going up since (-) and (-) should give a + (my reasoning was that if genes associated with negative regulation of immune response are getting down regulated then that means that the response is going up). But after thinking some more I simply don't know what to think about it at all. Now I do understand the meaning of ES/NES score and that all GSE does is telling me that genes associate with that term are at the bottom of my ordered list so they're downregulated and that genes may be common across GO terms, but I still didn't expect such contradicting things, so I must certainly not understand something. Any help?

Answer 1

[i.sudbery]

The thing to understand about enrichment analysis is that it doesn't tell you about the activity of different pathways, it tells you about the expression of the genes in those pathways, and the two things are not the same. You cannot conclude that because their is an enrichment for genes in a pathway, that the activity of that pathway has changed. It might suggest this as a hypothesis to go test, but it definately can't be taken as proof.

Secondly dividing things into "positive" and "negative" regulators of something, particularly something as broad as "immune response" is not very meaningful, as plenty of genes can be both positive and negative regulators in different circumstances.

Also remember that many of these things are members of signalling pathways, and it its the conduction of the signal, usually via post-translational modificaitons, that actaully has the biological effect.

I don't know specifically what your experimental design is, but for example, a T cell is always going to express more "immune response" genes, both positive and negative, than, say, a neuron. This also applies to more similar cell types, but I do not have sufficient knowledge of immunology to give particualar examples.

Finally, feedback loops are common in immunological regulatory pathways, and activation of an immune signalling pathway often leads to upregulation of the negative pathway to limit the time span of the reaction.

So where does this leave you? More often than not enrichment analysis is a hypothesis generation aid. If you were interested, I'd probably get hold of the genes in the leading edge and start doing some reading about them.