sourmash tax question - genus level matches?

[jodyphelan]

If a hypothetical new species was sequenced which had max 85% with all other species in the genus, would it be possible with sourmash to identify that it belonged to that genus?

To give a more concrete example, this is the output from gather with gtdb-rs207.genomic-reps.dna.k31.zip:

overlap     p_query p_match avg_abund
---------   ------- ------- ---------
97.0 kbp       1.5%    1.6%      38.3    GCF_000266905.1 Mycolicibacterium chubuense NBB4 strain=NBB4, ASM26690v1

[ctb]

hi @jodyphelan, some hot takes per @bluegenes (who wrote this part of tax):

that's complicated. Basically yes, if we believe we have the "right" ANI/AAI cutoff for genus-level matching

You might also consider using k=21 to get better genus level matching and/or use a protein molecule type/sketch (but we don't provide standardized protein databases yet, so you'd have to sketch your own matching genomes).

[jodyphelan]

Thanks @ctb

I've managed to boost the proportion in Mycobacterium to upt 11% by making k=21

sample name    proportion   cANI   lineage
-----------    ----------   ----   -------
WMW1132           88.9%     -      unclassified
WMW1132           11.1%     87.8%  d__Bacteria;p__Actinomycetota;c__Actinomycetes;o__Mycobacteriales;f__Mycobacteriaceae;g__Mycobacterium

[ctb]

@jodyphelan this looks great! I will say that an 11.1% overlap in k=21 space between microbial genomes is actually surprisingly stringent (despite looking "low") and would match ~family or genus level, as the cANI suggests.