Suppose to perform GSEA. Usually from an ontology file and counting occurrences of GO categories in the interesting set and the background we end up with p-values for the enrichment/depletion of the different categories. However, in bibliography different ways to do this are available. One can for instance decide to propagate each annotation to include all its parents, as it is clear that if you are annotated in a category which is child of another one, you also automatically belong to the most comprehensive one. However, I don't know of any paper where this difference has been verified to see if it improves or not the output. Anyone knows papers where this issue has been explored? what are your ideas and feeling in this respect?
I haven't tried running the GSEA algorithm on GO annotations. However, since GO is a DAG, some people have been successful with weighting child and parent terms to get a more biologically-relevant p-value. One paper that applies such a method is here: http://bioinformatics.oxfordjournals.org/content/26/7/905.full.
In addition, some genes have been annotated much better than others, just because not all genes have been researched equally.
The issue has also been addressed in:
Bioinformatics. 2007 Nov 15;23(22):3024-31.
"Improved detection of overrepresentation of Gene-Ontology annotations with parent child analysis."
Grossmann S, Bauer S, Robinson PN, Vingron M.
I would not use GO collection. It will be difficult to get a clear answer to your question. KEGG, reactome and Biocarta are good compromise. Hallmark is useful for cancer studies. Good luck in your analysis.
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology