I did feature-classifier classify-consensus-vsearch using my data and identified the assigned species.
And I checked the OTU in the qzv file by blasting it, and it was confirmed that the species was assigned to a low maxscore.
If per_identity is set high in the vsearch process, this problem disappears, but I wrote because I was curious about the reason.
You are using different databases, so might not get the same exact answer.
I am not 100% what the problem is based on the info you gave, but here are some ideas for you to check.
It sounds like this issue could be due to a misannotation in one database or the other. Unfortunately, many databases do contain mistakes, and this could explain why you are seeing hits to Anopheles with one database/method and Paraprionospio with the other.
I would guess that the error is in the database that you are using offline — setting a low percent identity threshold is probably causing a hit to the Anopheles sequence. Especially if you set --p-maxaccepts too low. Setting a high percent identity threshold causes that sequence to be excluded. The parameter settings might need some adjustment to work well for insect COI sequences (these settings were optimized using bacterial and fungal sequences).
You could visualize the alignment to both the Anopheles top hit (with vsearch) and Paraprionospio top hit (with blast) to visually assess the alignment quality...
Anyway, just some ideas, let us know what you find!
