search

nextstrain / avian-flu

Nextstrain build for avian influenza viruses
http://nextstrain.org/avian-flu
19 stars 7 forks source link

Reduce inference in cattle outbreak #66

Closed joverlee521 closed 3 months ago

joverlee521 commented 3 months ago

Description of proposed changes

Resolves #65 by

  1. setting divergence as default view
  2. ~removing augur traits so stop inference of location~

Checklist

trvrb commented 3 months ago

@joverlee521: I don't want to completely remove ancestral trait inference. The Texas origin is important. I'm working on a strategy here that will keep inference for known locations but won't infer locations for SRA sequences. Could you revert 828f97725a200734d228fd196dedb43576cc12f6 and I'll append to this PR?

trvrb commented 3 months ago

I tried a couple different approaches before landing on the commit above:

  1. Modify the traits rule to produce an inferred labeled as division_inferred. Update Auspice config to have separate division (direct from metadata) and division_inferred (via traits) colorings. However, because lat/longs don't have division_inferred this made map not work appropriately.

  2. Modify the traits rule to swap ? in the input metadata to Unknown location. However, this was causing funkiness where deep internal states were assigned as Unknown location due to there being so many tips with Unknown location. It's possible that --weights could have corrected for this but I decided not to explore further.

Instead I went with:

  1. Add a heavy amount of uncertainty via --sampling-bias-correction. This causes A/CATTLE/USA/24-013021-002/2024 to go from 68% to 20% confident in Ohio location, while leaving the deeper nodes adjacent to resolved GenBank samples to remain confident.

As expressed in commit message, this revealed a bug in Auspice in which Auspice currently colors tips according to most likely state unlike the logic it has for internal branches which are grayed out as entropy in state confidence increases. However, I think this commit is still appropriate to merge in avian-flu even if we're waiting on Auspice update to properly display.

We could still decide to implement (1) in addition to (3) to get an observed only geo coloring, but I'd like to see how the Auspice fix looks before doing so.

trvrb commented 3 months ago

I believe we can merge this PR as it stands, update live build and wait for Auspice update before doing further work here.