So I am not sure about the advantage of performing the differential abundance analysis at the genus level. Can you explain more? Thanks in advance.
Hi @jiakang_yin,
Reasons for doing the testing at the genus level would include (1) that it can be faster if you have a lot of ASVs, though that's not a great reason to restrict your analysis to less specific taxonomic levels, or (2) because you suspect that differences below the genus level are unimportant, which (in my opinion) often isn't likely.
Because of both of the caveats that I mentioned here, I typically do differential abundance testing at the ASV level, not at the level of a specific taxonomy.
Hope this helps!
Thanks for your reply.
Hi @jiakang_yin and @gregcaporaso,
Lately, I have been thinking on this topic. I think that 16S amplicon sequencing is usually considered reliable at the family or genus level due to most databases don't curate at species level. So, though it is interesting to identify which ASVs are driving the differential abundance results, could this lack of curation at species level be a possible reason to collapse at higher taxonomyc levels (eg. family or genus)? Maybe the differentially abundant genus would be more reliable?
Thanks!
Hello @andresarroyo
Maybe? There is a risk of combining ASVs though. As @jwdebelius found here, merging ASVs based on taxonomy means you are losing within-genus resolution; the database is holding you back!
One major challenge for in silico validation is the limited resolution of existing databases; our results exceed the OTU-based resolution used in construction ...
Maybe? You can find the sum, mean, or median of counts from ASV1 and ASV2. I suppose reporting a median (calculated across one or more features) would be less sensitive to outliers But if only one ASV is changing between groups, this signal would be reduced.
Conversely, Pat Schloss (PI of the Mothur project), found that OTUs can be better than ASVs in specific situations and document that here:
Note, that they are making OTUs without a database, so it's also unbiased by the database, unlike collapsing/merging ASVs based on predicted taxonomy.
It's a trade-off.
Thank you so much for your response, @colinbrislawn! As suggested in other posts, I think I will develop the analysis at the ASV level and other taxonomic levels to gather as much information as possible and to be able to compare results.