Can we further tune the outlier filtering to retain true type strain records?

[marykstewart]

Hi Chris and Noah, last year we exempted type strain records from outlier filtering, which has cleaned up many blasts. Now I'm collecting the instances where legitimate type strain records are being excluded, and trying to learn from these instances how the logic could be further modified toward the goal of including true type records and filtering incorrectly classified type records.

Here are 3 examples:

• Faecalibacterium longum https://gitlab.labmed.uw.edu/molmicro/uwnt-management/-/issues/359

  • 5 records total in ya16Sdb, max dist 1.11
  • 1 type strain record, which is the only outlier

• Segatella baroniae https://gitlab.labmed.uw.edu/molmicro/uwnt-management/-/issues/361

  • 36 records total in ya16Sdb, max dist 18.07
  • 5 type strain records, all excluded
  • For this example, I think that excluding records from whole genomes with FAILED tax check status would right the ship. The criteria for failing tax check (as opposed to receiving a status of INCONCLUSIVE) are rather stringent. The ANIb of the query genome compared with the type genome of its designated species has to be BELOW threshold, AND the ANIb of the query genome compared with the type genome of a different species has to be ABOVE threshold.

• Corynebacterium pseudogenitalium https://gitlab.labmed.uw.edu/molmicro/uwnt-management/-/issues/371

  • 19 records total in ya16Sdb, max dist 7.69
  • 4 type strain records, all excluded and all from true type genome https://www.ncbi.nlm.nih.gov/nuccore/NZ_CP072934.1
  • I'm not at all surprised to see that this taxon makes my list, since the taxonomy here is historically complicated. Corynebacterium pseudogenitalium was proposed in 1979 but not fully described until recently, see https://www.sciencedirect.com/science/article/pii/S0923250822000687?via%3Dihub, and the winding history of the taxon has left a trail of records that are no longer considered to belong to the species.

I think excluding genomes that fail taxonomy check is low risk and that if possible we should implement that. This implementation would also clean up some other blasts, for example excluding LSCY01000004 and LSCY01000011 here.

For all 3 of these instances, all type strain records in the db end up in the outlier pile. Can we lean on that as a criterion to exempt a taxon from type strain outlier filtering?