brainstorm how to evaluate the impact of "bad" databases/taxonomy on performance

[ctb]

we are using GTDB taxonomy because -

• GTDB provides a nice collection of representative genomes
• GTDB provides species clusters based on ANI, which matches the approach used in charcoal
• the genomes used by GTDB do not seem to be terribly contaminated (viz sourmash-oddify)

we are not using NCBI because:

• it doesn't provided a versioned or stable taxonomy
• we'd have to do our own clustering
• initial explorations with sourmash-oddify suggested that there were many taxonomic discrepancies (vs ANI/shared blocks of sequence) present in NCBI taxonomy

but it would be good to evaluate all of this more clearly.

one thought might be to find and characterize "high rank" k-mers (k-mers above match rank; see e.g. this blog post) from the database. sourmash_databases has some code to do this.

[ctb]

36 is potentially relevant, in terms of doing some kind of cross-MAG analysis within charcoal.

[ctb]

continuing down the rabbit hole of sourmash-oddify and high-rank k-mers, it might be interesting to see if we can quantify the impact of charcoal decontam by looking at what kinds of high-rank k-mers they would have introduced into the databases.