Hi All,
We’re analyzing mouse fecal samples using our shotgun data. We specifically want to test whether a specific taxon contributes to a particular functional pathway.
We ran HUMAnN 3.9 with ChocoPhlAn v31 and UniRef90. We have two groups. We ran MaAsLin using MetaPhlAn output and found that the taxon we’re interested in is more abundant in one of the groups. However, HUMAnN is not attributing the pathway to that taxon.
What would be some potential next steps? We’ve been running HUMAnN following the manual. Should we consider re-running HUMAnN with different settings or a subsetted reference database? We’ve done some post-run checks: HUMAnN could assign many gene families (and many pathways) to the family we’re interested in, but it doesn’t have enough evidence to characterize the pathway.
Thanks!
Claire
Does the taxon of interest encode the pathway of interest, and, in particular, in the humann database? This can be tough to discern, but from what I can tell if a taxon encodes a function/pathway then humann outputs a 0 (rather than an absence or NA) for that function and so you should be able to see that it is present in the database (line in the tables) but not in your data (0 values for depth). Be aware also that there are several pathway databases (kegg, metacyc), and these will aggregate pathways differently, so perhaps a test of running humann with the other pathway reference may help. Other options to check include, for example, the literature, in mouse/fecal derived genomes (isolates, MAGs, etc).
If there is no evidence anywhere for the taxon to encode the pathway, then no amount of running humann will support the hypothesis because the pathway does not exist for the taxon in humann’s database.
If there is in fact a possibility that the taxon encodes the pathway, then there is a likelihood that it does not in your dataset, ie gene flow (loss) compared to the humann references. In this case, lack of presence of the pathway may suggest that it is missing in your taxon, but certainly does not confirm. Human and mouse fecal communities are very different at strain granularity (even different humans and different mice), which humann is arguably not designed to resolve.
Again, no amount of running humann will likely change the outcome.
Note that I am not a dev, purely speaking from experience.
I would strongly recommend upgrading to MetaPhlAn 4 + HUMAnN 4 as a start as they are much, much better at assigning functions to mouse-associated communities compared with v3 (as discussed in the MetaPhlAn 4 paper).