Non-homogenous one per branch model likelihood computation

[Keren Halabi]

Dear Bio++ team,

I have a difficulty understanding the logic of likelihood computation of non-homogeneous models , with one_per_branch setting.

According to the bppSuite manual, page 23: "When nonhomogeneity option is one_per_branch, each site is constrained to follow the same submodel from leaves to root."

According to this logic, I would expect that the likelihood computation (before any optimization) for a non-homogenous model with one_per_branch setting, where the model settings are or M2 with set of parameters values X, would yield the same likelihood as the site model M2 with the same set of parameters values X.

Unfortunately, when testing this assumption on each setting produced different log likelihood value.

I attach here the parameter files used for each option (one_per_branch and homogenous_site). Both used the lysosome data provided in your Examples folder.

The site model yields likelihood: -1072.277223274

The non-homogenous one_per_branch model yields likelihood: -1077.8222057178

Could you please clarify why that is?

Many thanks!

Keren

[Laurent Guéguen]

Hi Keren,

I am sorry for the mislead, but this description is false; it is the remain of a previous implementation.

In one_per_branch setting, each model on each branch is independent,  so the mixture in M2 is done branch specifically

(in the same manner it is done in hyphy, actually).

I fix the manual, sorry.

Cheers,

Laurent