GO and KEGG

Question

[matteo.levorato]

Hello,

I have some questions about GO and KEGG gsea in Rstudio:

• which should be the input genelist for the function gse? at the moment I am using this:

    geneList <- geneList[abs(dataGSEA$log2FoldChange) > 1 & (dataGSEA$padj) < 0.05]
  

And then I sort the samples

geneList = sort(geneList, decreasing = TRUE)

Is this correct? should I use only the up-regulated ones?

Moreover, I'm still confused about the difference between GSEA and ORA analysis (from a practical point of view):

• when should I use ORA instead of GSEA? Should I use both?
• for the enrich function should I filter the gene (by abs log2foldchange and padj) like for the gse analysis?

And last question:

• which should be the input list for the fgsea function and it should be sort?

Thank you very much in advance

Answer 1

[jv]

there are numerous resources online, including this forum, to help answer these questions. Have a look at the following:

• https://www.gsea-msigdb.org/gsea/doc/GSEAUserGuideFrame.html
• https://yulab-smu.top/biomedical-knowledge-mining-book/enrichment-overview.html
• https://www.biostars.org/post/search/?query=GSEA+rank

Moreover, I'm still confused about the difference between GSEA and ORA analysis (from a practical point of view):

Personally I avoid ORA whenever possible, it's a flawed method in that it's very dependent on how you select your genes of interest and at times it can be tricky to select an appropriate background set of genes. With GSEA you use the complete differential gene expression results so there should be less (human) bias and more information to derive meaningful results.

[matteo.levorato]

I red the guide before post and after your comment, but I swear that for me is not clear if I have to apply any filtering steps of my gene list before perform ORA and/or GSEA

[jv]

for ORA yes, you need to select a subset of genes of interest. Some examples include statistically significant upregulated genes or a cluster of co-expressed genes.

For GSEA you do not filter your ranked gene list though there may be some pre-filtering of lowly expressed genes before doing something like a differential expression test for example.

[matteo.levorato]

Thank you very much!!

Can you explain me better the pre-filtering step?

[jv]

https://bioconductor.org/packages/release/workflows/vignettes/rnaseqGene/inst/doc/rnaseqGene.html#pre-filtering-the-dataset

[jv]

if my answer was helpful please upvote and accept.