Hello Biostars
I´ve just started using GO enrichment analysis. I have ran a GO enrichment analysis using Blast2GO on a set of ~ 400 genes. One rather big group were assigned the GO term GO:0044260:cellular macromolecule metabolic process. This is quite a wide definition, and my question is as follows: does it makes sense to run a GO enrichment test on the genes with this term. Do I get a better resolution when it comes to biological function.
To be honest, I have already done this, and what I get is, as I hoped for, more narrow groups of different forms of metabolic processes. My question is still, based on how the analysis is performed, can I do this?
Thanks in advance
•
written
by
linelr •
It makes sense if you specify an appropriate backgroundlist/genome (not sure if you can do that in blast2go). Enrichment analysis will check if the genes in your list are more often found in a certain geneset than expected by chance alone, but if you limit the inputdata by making an artificial selection you are cheating the program, because it shouldn't compare against likelihoods of finding an enrichment for an entire set versus the entire genome, but just for the subset you included in the analysis.
Thank you for your answer
I do have a background list/ genome. It was used for the initial GO enrichment analysis. If I understand you correctly, my strategy is ok when I´m taking a subset of all the expressed genes (in this case the already analyzed and GO-associated genes) and compare them with my background list?
You can't make a selection based on enrichment and put it again in an enrichment analysis and expect a statistically sound result... Perhaps you could have a look at something else besides GO to give you clues, e.g. http://amp.pharm.mssm.edu/Enrichr/, which besides GO and KEGG includes multiple other pathway/geneset databases.
To deal with the redundancy of GO annotations that often leads to high level terms being enriched, I used the method in this paper. Another strategy I use is to only look for enrichment using a subset of GO (e.g. GOslim) that is relevant to the biology I am interested in.
Log in to answer this question.
What I usually do in case I get too many 'general description' terms is to filter out terms after the enrichment that have too many genes (or too little). I do this selection on the total numbers of genes per term (so not the significant number of genes in the term).
I don't know if this is cheating, but it helps to focus on 'useful' terms to get insight in your data. This way I filter out the too general and too specific terms.
Thank you for your answer
Sounds like a good idea. I´m not sure if I can apply it to my dataset, as I´m not lucky enough to have a lot of GO-enriched terms. They´re all quite broad.