The clade I builds were using the clade IIb reference which is far away.
Switching to a clade I reference makes the trees better.
I curated the mask.bed file by looking at the new alignment, with an eye specifically on fixing clade Ib artefacts. I masked a couple of suspicious regions where there was a mutation right next to a stretch of Ns in sequences of one institution only (usually INRB).
New tree is much more sensible than the old one. The old one had sequences cluster per sequencing institution (INRB vs Erasmus), the new one mixes them nicely as it should.
The clade I builds were using the clade IIb reference which is far away.
Switching to a clade I reference makes the trees better.
I curated the mask.bed file by looking at the new alignment, with an eye specifically on fixing clade Ib artefacts. I masked a couple of suspicious regions where there was a mutation right next to a stretch of Ns in sequences of one institution only (usually INRB).
New tree is much more sensible than the old one. The old one had sequences cluster per sequencing institution (INRB vs Erasmus), the new one mixes them nicely as it should.
New build example: https://next.nextstrain.org/staging/mpox/clade-i?c=num_date&label=clade:Ib&m=div
Compare to currently live build which is unrealistic:
https://next.nextstrain.org/staging/mpox/clade-i?c=num_date&label=clade:Ib&m=div
Now one can actually see a time signal of sorts and the root isn't entirely unplausible.
I added a bunch of colorings to allow better investigation of sequencing-institution-specific artefacts/clustering or absence thereof.
We should probably turn off timetree, it's just too wrong, in particular for clade Ib.