ClusterProfiler-enrichGO questions about the universe

Hi there,

II have been analysing my RNA-seq data for a short time and don't have much prior experience with it. I have taken some courses to learn the basics, but I have a few questions about the clusterProfiler package, specifically the enrichGO function.

First of all, sorry if I have some misconceptions or if something is not well said, but I have no experience at all before this study.

My data came from mouse liver mRNA. What I have read is that it is better to define a proper universe with all the genes that were detected after DESeq2. However, some of my colleagues who work with tumoral pancreatic cell lines told me that they didn't define any universe; they just took the whole, in their case, human genome. I understand that the scenarios are not the same, but that makes me hesitate about whether I'm doing it properly by defining a universe.

In addition to that, what I wanted next was to analyse possible GOs and pathways that may be enriched in the resulting genes from a Venn Diagram. The thing is that I have done a Venn diagram, and with those genes that are exclusive to each condition, I did an enrichment analysis. In this case, I have defined the universe as the union of all the DESeq2 genes of each condition. However, the results showed GO terms that I don't think are appropriate for the liver, like cochlea development. To my mind, this is because those genes are maybe transcriptional factors that may be involved in multiple pathways, but maybe I have been using the wrong approximation for the case.

So, if anyone could help me, I'd be grateful.

Esperanza

enrichgo clusterprofiler

• 84 views • link

updated 2 days ago by jared.andrews07

• written 7 days ago by Esperanza •

1 answer

See here. In short, yes, providing the background is important and will change results meaningfully.

However, some of my colleagues who work with tumoral pancreatic cell lines told me that they didn't define any universe; they just took the whole, in their case, human genome. I understand that the scenarios are not the same, but that makes me hesitate about whether I'm doing it properly by defining a universe.

They're wrong, you should always set the universe/background. People do stuff wrong all the time (see post-hoc filtering of DE genes for another prime example).

Your analysis approach for the second part of your question sounds appropriate to me.

However, the results showed GO terms that I don't think are appropriate for the liver, like cochlea development. To my mind, this is because those genes are maybe transcriptional factors that may be involved in multiple pathways, but maybe I have been using the wrong approximation for the case.

GO terms can be hilariously broad, are largely associative rather than causal, and can have very heavy overlap. You will nearly always have significant terms that seem irrelevant or inappropriate. Semantic similarity analyses can help collapse these terms based on overlap to keywords/broader terms that may be more informative or appropriate. Most of the time, people cherry pick the hell out of these results to show what they find most interesting/novel anyhow.

• 0 views • link

written 7 days ago by jared.andrews07

Thank you so much for your reply.

I just have another question related to using a universe for EnrichrGO. The OrgDb is Mus musculus and with a set list of 219 and a universe of 20764. I obtained no GO enrichment for Biological Process or Molecular Function, but a few for Cellular Component. Therefore, I was hesitating about whether to define the universe or not in light of the results

• 0 views • link

written 3 days ago by Esperanza •

My advice stands. Decreasing thresholds/stringency may be appropriate in certain contexts, but doing so just to get "significant" results is typically not a great idea. p-hacking is to be avoided.

• 0 views • link

written 2 days ago by jared.andrews07