How Can Pathway Analysis Methods Account for Genes with Different Activation Roles?

Question

[RoyBatty]

I've been using pathway analysis methods like GSEA and GSVA, which evaluate pathway activity based on gene expression levels. However, these methods seem to assume that the more highly expressed a gene is, the more it contributes to pathway activation. This seems to oversimplify how many biological pathways work, where genes within the same pathway can have different activation requirements:

• Some genes (e.g., activators) need to be upregulated for a pathway to be active.
• Others (e.g., inhibitors or regulatory genes) need to be downregulated or remain at baseline levels for proper activation.

Given this, I’m wondering:

1. Are there any methods or modifications to GSEA or GSVA that account for the fact that some genes in a pathway may need to be downregulated or have specific expected activation levels for the pathway to function properly?
2. Alternatively, are there tools that incorporate the biological roles of genes (activators vs. inhibitors) into pathway analysis, rather than just relying on overall expression levels?

Also:

1. Do public databases like KEGG contain information on the gene role within a pathway? For example, this gene should be highly expressed, or lowly expressed?

I’m curious how others approach this issue and whether there are tools or methods that explicitly address it. Any insights or suggestions would be greatly appreciated!

Answer 1

[i.sudbery]

which evaluate pathway activity based on gene expression levels

This is your mistake. They do not do this. It is a common misconception, but what tools like GSEA give you is an idea of the themes in your gene list. They tell you "this gene list contains more genes associated with pathwayX or cellular component Y, or with biochemical activity Z, than we'd expect". Whats more, the same genes when differential in small gene lists might be enriched, but not when there is a big gene list.

For example, if a simple pathway has 3 genes, and they are all upregulated. If only 10 genes in total were upregulated, then enrichment tools will probably find this pathway is enrichmend. If 10,000 gene were upregulated, then this pathway won't be enriched. But the change in all three genes in the pathway will be the same in these two cases. If all genes were twice as active in treatment vs control, if doesn't really matter what happens in the rest of the genome to know if the activity of that pathway has changed.

Whats more, it these genes are signalling pathways, then the "activation" of the pathway is far more dependent on the strenght of the chemical signal the cell is recieving, than the protein expression levels of the genes in the pathway (which is anyway only partially correlated with the RNA level) .

These tool simply give you the name of a pathway you might think is intereting to look at more closely. They do not give you a biological interpretation, like pathway activation etc. That does of interpretation can really only be given by a human expert interpreter. Best to think of them as hypothesis generating tools.

There are tools that try to do this sort of thing. The one I have come across is SPIA. But I didn't have a particularly good time with it, because it depends heavily on the pathway annotation fed to it, which is often incomplete or hard to interpret. KEGG does annotate connections between factors as activating or represssing in some cases, but in other the relationship is things like "binds to", which can't be interpreted, and causes breakages in the pathway.

Another thing you can do if you are interested in signalling pathways is to look at the downstream effect genes that form a signature of activation by that pathway.

Answer 2

[carolofharvest]

Yes.GSEA or GSVA pathway analysis only show there is enrichment in those pathways but we can not know those genes causing enrichment are supressors or activators of that pathways. Some pathways have indications like "up" and "down" though.

I know progeny pathway analysis from decoupleR package. Each gene has a weight score in that database.

I would like to hear others suggestions about this too.