Pathway Enrichment With Expression Values

Question

[Dataminer]

Hi!

I know there are very nice tools like DAVID, GSEA, WikiPathways etc, for pathway enrichment analysis. Where they take bunch of genes as an input and produce a list of enriched pathways with some p-values.

My questions is, what if I have affymetrix probe ids, gene name and the expression value of gene in two cell types (say CD34+ and leukemic), and I want to look for the pathways enriched by these genes based on fold change between different conditions or cell types.

The data is in the format of log2 and the organism is human.

Could you please suggest a tool or a method to solve such a puzzle.

Thank you

Answer 1

[ff.cc.cc]

Hi, It seems that you still have not analysed your data, but only log2 transformed. If this is the case you should :

• normalize the dataset (there are many R bioconductor packages: affy & c.)
• compute fold changes (e.g. limma R package and its good tutorial)
• compute enrichment with one of the tools you listed, based on toptable() results

hope could help

p.s.

there are other threads to go in deep with these topics (tags: limma, affy, quantile normalization, vsn... )

[Dataminer]

Hi! I have already done the pre-processing of data (as suggested by your comment. I know the set of genes going up and going down (differentially expressed) between two cell types/conditions. What iw ant to know how can I use this information (gene along with expression values) to look for the enriched pathways. I hope i have cleared myself.

[ff.cc.cc]

I understand you want to capitalize on the results of differential expression tests, brainstorming suggestions: i) (as above) use the fold change info only to select top differentially expressed genes and put them into GSEA (or others) ii) as i), but compare your list only against other gene expression signatures, but difficult to get also the same conditions/phenotypes iii) hack the R code of GSEA (that seems documented and well written) to take into account an additional weight (a value related to your fold change) when computing its score iv) skip enrichment and look at meta-analyses with similar experiments

Answer 2

[Dataminer]

Well it is not always nice to first ask a question and then come up with an answer. Anyway the answer to my own question is SPIA package from R here it is. Rest I will leave on you all to dig and discover ;) ...