I have some sequencing data and I identified that a specific genus is more highly abundant in a group versus another one. For this, I used R and Lefse based on a biom file (created after merging the feature table with the taxonomy file).
In order to identify all ASVs belonging to this genus, I merged the repseq.qza with the taxonomy file. From there, I filtered all ASVs (n=16) related to the genus of interest and used blast to obtain a more detailed taxonomic information for each one of those 16 ASVs.
My questions are:
According to ANCOM (using a not collapsed feature-table), out of those 16 ASVs, only one ASV was significantly different between my groups of interest. So, this means that the other 15 ASVs have a similar abundance between both groups? And that my relative abundance comparison results (for the same genus) are only being driven for one specific ASV?
Besides ANCOM, is there any other way of comparing the relative abundance of specific ASVs (related to a specific genus) between 2 or more groups?
I checked the forum, and another user had a similar question.
Its definately possible that only one ASV in a genus is driving a pattern. It’s not uncommon, persay? (Although I can’t give you exact numbers; there are definately a lot more examples of a genus driving a larger classification, though, so the principle holds.)
It may be worth keeping in mind that a genus can be a broad functional classification and that while you’re doing molecular fingerprinting, there’s a lot of information you’re not seeing. (I just discovered my house cat is in the same genus as an African wild cat. I’ll have to remind her that next time she freaks out about me bringing home flowers ).
I would use your volcano plot to look at this. You should be able to (I think) find the 15 ASVs in the distribution. I’d check both the W value and the CLR transform.
In terms fo relative abundance, you could generate a heatmap off a filtered table in QIIME or in R (depending on where you think you’ll have the most control) or some kind of barchart.
Finally, if your data is already in R, you might consider Phylofactor which is an isometric log transform function that partitions your data based on a phylogenetic tree. It’s similar to Gneiss, but slightly different in that you can get polyphyletic clades, where it will separate out a tip in the middle of your tree. (Sadly there is no QIIME 2 implementation). It might be worth seeing if your single ASV replicates or if it gets pulled out with the whole genus, or if there’s something nested.