GO Analysis Combine or Separate DEGs

Question

[Koki]

Hello,

I know there have been previous threads on this, but I'm making a new post to see if the consensus has changed over the past few years and want to ask some additional questions.

I am performing bulk RNA-seq analysis of immune cells and expect to see an enrichment of immune related terms between my conditions. However, separating DEGs and performing GO analysis brings up enrichment of the same terms in both the downregulated and upregulated sets. I'm not sure how to infer any directional change from this data.

Additionally, I want to understand the nuance whether to separate or combine DEGs. I'm guessing it goes something like this:

1. Separate Analysis

• You have a hypothesis about the direction you expect GO terms to change

1. Combined Analysis

• Exploratory, want to see what terms change without an idea of directionality

I am conducting the analysis using clusterProfiler. Any clarifications would be appreciated!

Answer 1

[mark.ziemann]

performing GO analysis brings up enrichment of the same terms in both the downregulated and upregulated sets. I'm not sure how to infer any directional change from this data.

This happens when a background list isn't being used. See this article for more info. When running ORA analysis a background is critical, forgetting this step can cause up to 40% of results to be false.

In general, separate analysis is better because genes in a set are typically possitively corrrelated, so you should see distinct pathways up and down which will give you insights as to what cells/tissues are doing. There was a paper on this a while back.

Better still, use GSEA or other FCS approaches which are more sensitive. Evidence here, specifically figures 3 and 4. When using that type of analysis, it includes all detected genes, so you will not be affected by the background problem. Clusterprofiler has a GSEA() function which could be the most convenient for you.