Have you been using a file
trait_weights.csv to address the impact of sampling bias during ncov ancestral trait reconstructions using
augur trait (as discussed in the github page below)?
Could you point me to a Snakefile on github where the
--weights command is used?
We played with this in the zika build and in early ncov builds, but it never really improved things too much. I don’t think it is currently used in our analysis. What it does is constrain the equilibrium frequencies of the GTR model used to infer ancestral states.
You can check out the
ncov repo in ~March for how this worked: https://github.com/nextstrain/ncov/blob/bab177d26f9dad7f2a6e1dec488952f039fefadc/Snakefile#L233 with weights file: https://github.com/nextstrain/ncov/blob/bab177d26f9dad7f2a6e1dec488952f039fefadc/config/weights.tsv. At the time we were significantly under-sampling local Chinese cases compared to travel exports. Adding weight to these under-sampled regions definitely helped the mugration model to give more accurate results. (You can do this in BEAST as well when setting equilibrium frequencies)
Some simulations would be in order, but I believe if you had case count data indicating truer burden across geographies, you could tune
--weights in a more empirical fashion and improve mugration inference.
@rneher and @trvrb, thank you for your feedback.
I’ll try both
augur trait and
BEAST approaches and compare the outcomes.