I want to compare pathway abundances generated using HUMAnN2, by LEfSe.
I merged tables across samples, and performed humann2_split_table, and now I have a raw table, a stratified table, and an unstratified table.
From my understanding, I should input a raw table to LEfSe, as in the MetaPhlAn2 tutorial they input a table which preserves the rank (not summarizing to a certain rank). However, I saw some websites and papers that input an unstratified table.
What input is desirable to perform differential analysis? Is it depending on the purpose of analysis?
I would use the unstratified table. The stratification support in LEfSe is geared toward taxonomic stratification (species within genera within families etc.). While the stratification in HUMAnN2 looks similar it has a different meaning.
Hello franzosa, thank you very much for your reply.
Could you please clarify differences of stratification in HUMAnN2? My specific question is that “Are pathways contributed by the species we are interested different between groups?”. In this case, is using LEfSe to the stratified output table limited to species we are interested okay?
In the wiki of HUMAnN2, there stated
Unlike gene family abundance, stratified pathway abundance does not necessarily sum to yield the community total pathway abundance. Why not? (Hint: HUMAnN2 estimates the number of complete pathways in the community or a given stratification.)
Whether they are not summed to one or not, comparing these abundances (stratified abundances of limited species) between groups is I think not a problem. Is this wrong?
I see - if you’re specifically interested in the abundances of pathways in specific species, then working with the stratified table makes sense, and comparing stratified abundances between groups is definitely not wrong.
My concern was that there might be something baked into the LEfSe software assuming that stratifications = taxonomy (since LEfSe predates HUMAnN 2.0 by several years), but that’s probably unlikely. If you encounter any weird behavior let us know!