why use TN93 mode not others?

[liamxg]

Is this the best mode for genetic distance? @stevenweaver @spond

[spond]

Dear @liamxg,

There is no best genetic distance. The TN93 distance has a simple closed-form expression, i.e. you can compute it from pairwise counts. You can't do it for more complex models (e.g. REV). Simpler models, like K2P, are poor choices for HIV-1, because it has strong compositional biases (e.g. A ~ 40%). One reference that I recall is https://pubmed.ncbi.nlm.nih.gov/15750402/

For small distances (e.g. <1%), there is generally little difference between the actual model used, but we might as well keep a realistic model in place.

Best, Sergei

[liamxg]

Thank for your reply, could we do model selection first for each dataset, and use the optimal model to calculate the genetic distance?