High Percentage of Unmapped and Unintegrated Reads in HUMAnN3 Analysis

First, I want to extend a big thank you @ franzosa to the developers for your amazing work in creating such a powerful tool HUMAnN3!
I recently used the following code while analyzing a colitis mouse feces sample with HUMAnN3.
humann -i ./${i}_1.fq.gz -i ./${i}_2.fq.gz -o ./${i} --threads 24 --memory-use maximum --search-mode uniref90
HUMAnN3 itself went very smoothly, but I’m struggling to make sense of the results. Specifically, I’m seeing close to 99% of my reads mapped as UNMAPPED and UNINTEGRATED, which is quite concerning. I understand that the proportion of reads mapped directly to pathways might be lower, but having nearly 99% unmapped seems unusually high and unacceptable especialy in feces samples.
Is there anything I can do to improve the mapping rate? Any insights or suggestions would be greatly appreciated!
Here is finnal result,
humann3_renom_pathabundance_unstratified.tsv (55.8 KB)

Additionally, if any suggestions on methods for analyzing differences between groups using the HUMAnN3 results, I’d really appreciate the guidance!

To test whether it is a sample/seq issue, you can try to use uniref50 and see if the mapping rate increases. Maaslin2 is a good R package to analyze HUMAnN3 results.

You can search here for several previous posts where I discuss options to improve mapping rate. It depends a lot on whether your rate is low because of low read depth (in which case you need to tone down coverage parameters) or high novelty (in which case mapping to UniRef50 will help). Switching to the newly released HUMAnN 4 may also help with the novelty issue, since the HUMAnN 4 database includes a lot of species assembled from non-human environments AND automatically falls back to UniRef50 for translated search.