Usual data:
Common questions:
The usual phylogenetic posterior:
$$P(T,\mu,\theta|A) = \frac{1}{P(A)} P(A|T,\mu)P(T|\theta)P(\mu)P(\theta)$$
Where does geography fit in?
Currently two main classes of models:
The usual phylogenetic posterior is:
$$P(T,\mu,\theta|A,L) = \frac{1}{P(A)} P(A|T,\mu) P(T|\theta)P(\mu)P(\theta)$$
where
The standard phylogenetic posterior is modified:
\begin{align} P(T,\mu,\theta|A,L) =& \frac{1}{P(A)P(L)} P(A|T,\mu)P(L|T,M)\\ &\times P(T|\theta)P(\mu)P(\theta) \end{align}
where
Note the similarity between the two tree likelihood terms.
Mugration models treat location as just another trait/character.
A very important assumption made by the mugration model posterior:
Samples are assumed to be collected in a manner that is blind to their location.
A helpful way to visualise the mugration model is to imagine its effect on the population as a whole:
Use a contiunous diffusion process in place of the discrete random walk:
Essential features of mugration model remain, including sensitivity to sampling.
Note that $m_{ij} = q_{ji}\frac{N_j}{N_i}$ where $q_{ji}$ is the forward-time migration rate from $j$ to $i$.
Again, the standard phylogenetic posterior is modified:
\begin{align} P(T,\mu,\theta|A,L) &= \frac{1}{P(A)} P(A|T,\mu)\\ &\times P(T,C|\vec{N},\bar{M},L)P(\mu)P(\theta) \end{align}
where
The sample locations and SC model affect the tree prior.
The shape of the tree is affected by structure.
The strucured coalescent makes no assumption about the manner in which samples are collected with respect to location.
Required packages:
- BEAST_CLASSIC
DensiTree output:
Required packages:
- GEO_SPHERE
Google Earth visualization example:
Required packages:
- MultiTypeTree
De Maio et al., PLoS Genetics, 2015
Required packages:
- BASTA
Comparisson between mugration implementation and full and approx. SC models.
Müller et al., MBE, 2017
Required packages:
- MASCOT
Required packages:
- MASTER
- MultiTypeTree
- SA
Time and space-dependent birth/death parameter estimation!