GSEA with Wikipathways not entirely reproducible
0
0
Entering edit mode
Susana • 0
@cb00165b
Last seen 7 weeks ago
Netherlands

In February I made a GSEA analysis using Clusterprofiler and WIkipathways and I got my results of pathways that are sign. enriched (p-adjusted0.05) in each of the conditions.. I wanted to confirm my results before publishing and re run the code recently. Everything is the same except 3 pathways do not come up anymore ( estrogen receptor pathway and two eicosanoid metabolism pathways) in any of the conditions it showed previously. Even if I ask to have all pathways enriched (put the pvalueCutoff=1) they do not appear. Which leads me to think that it is not they are not significant but something in the package perhaps changed? If I use the online tool Webgestalt and the same exact list the 3 pathways show up again ... Is there any explanation for three pathways suddenly not being considered by GSEA?

clusterProfiler • 97 views
1
Entering edit mode

Without having access to all your input files it is difficult to explain why this happened. However, some thoughts... Did you use the exact same version of R and Bioconductor packages as in February?

Thus:

• are you sure the input is identical? Thus the gene sets, and your list of input genes? Well, the importance of the former is obvious, but if gene annotation have been changed, this may affect the ID-to-gene mapping. Note that with each new Bioconductor release the gene annotations are also updated.
• By default only gene sets containing more that 10 genes are included in the analysis. A quick (manual) count showed me that the ER pathway only consists of 13 gene products, and none are entrez ids. clusterProfiler uses entrez ids as central input identifiers, and if the mapping entrez id <-> ensembl id has changed, this may result in having less than 10 genes mapped to the ER pathway.... See also my 1st comment.
• did you now that by setting a 'seed' you allow for reproducible analysis when random numbers are used. See e.g. here.