SetRank ranked analysis for pathway enrichment

Question

[elavelle]

I'm using SetRank with R for pathway enrichment analysis on tables output from DESeq2 (rows = gene IDs, columns = log2fold change, padj, ect.) The user manual instructs to create a reference set of genes for those with counts greater than zero across all samples (by using padj != NA on the comparison tables). The manual later states: "In ranked mode, the gene identifiers in the input vector are assumed to be ranked by increasing p-value." Making a vector of the union of genes expressed in all samples is simple enough, but how am I expected to order them for ranked analysis? For example, if in the comparison tables:

Condition 1 vs. 2

gene A 0.0023

gene B 0.0056

Condition 1 vs. 3

gene B 0.0011

gene A 0.0067

How should it be determined which gene is ranked higher in the input vector? Should I order each individually, then just get a loosely ordered vector with the "unique" function? Is it suitable to be combining the tables for all comparisons when there is more than one type of condition in the first place?

Thanks

Answer 1

[jobbe.goossens]

You should provide separate vectors for each contrast and run them separately. I don't know the details of your experiment but I suppose you want pairwise comparisons since you also got pairwise p-values for your independent genes. If this is not the case and you just want overal p-values than you should reanalyze the data on the single-gene level with a statistical test that does not output pairwise p-values but overal p-values f.e. two-way anova and use these for ranking.