Inconsistent taxonomic assignments with EPA-NG

[VeronicRojo]

Hi everyone,
I’ve been using the EPA-NG module along with the SBDI-GTDB database to perform phylogenetic placement of 16S ASVs (V3-V4 domains). I ran two independent analyses on two different batches of samples. While the taxonomic composition of the samples varies between batches, they all contain an inoculum of well-characterized bacterial strains. However, I’m getting different taxonomic assignment results for some of the inoculum strains.

For example, the same ASV (611027a5d9e7fb25bc2099e87451b10b.txt) was classified as follows in the output file sample.taxonomy.per_query_unique.tsv:

. First run:
d__Bacteria;p__Pseudomonadota;c__Alphaproteobacteria;o__Rhizobiales;f__Rhizobiaceae;g__Mesorhizobium;s__Mesorhizobium sp001427285 (LWR = 0.9952)

. Second run:
d__Bacteria;p__Pseudomonadota;c__Alphaproteobacteria;o__Rhizobiales;f__Rhizobiaceae (LWR = 0.9066)

When I review the more detailed sample.taxonomy.per_query.tsv file, it seems the second run had conflicting assignment at the genus level, so only the family-level assignment is reliable:

I'm puzzled by how the same ASV can end up with such different assignments across runs, especially since the EPA-NG parameters were identical in both cases. Is this variability a known issue with EPA-NG’s preplacement heuristics, or is there a parameter I can tweak to achieve more consistent results? Any insights would be greatly appreciated!

Thanks in advance!

[d4straub]

Any insight here @erikrikarddaniel ?

[erikrikarddaniel]

Any insight here @erikrikarddaniel ?

Actually not, although I assigned it to myself and then forgot, sorry for that.

I just checked the obvious -- if there's a param for random seed in epa-ng -- but found nothing.

Maybe you, @VeronicRojo , could ask at the RAxML forum: https://groups.google.com/g/raxml? (I could do it too, if you send the ASV, but probably quicker if you do.)