Prepare to run BiSSE (Binary State Speciation and Extinction) on a phylogenetic tree and character distribution. This function creates a likelihood function that can be used in maximum likelihood or Bayesian inference.
make.bisse(tree, states, unresolved=NULL, sampling.f=NULL, nt.extra=10,
strict=TRUE, control=list())
starting.point.bisse(tree, q.div=5, yule=FALSE)
An ultrametric bifurcating phylogenetic tree, in
ape “phylo” format.
A vector of character states, each of which must be 0 or
1, or NA if the state is unknown. This vector must have
names that correspond to the tip labels in the phylogenetic tree
(tree$tip.label). For tips
corresponding to unresolved clades, the state should be NA.
Unresolved clade information: see section below for structure.
Vector of length 2 with the estimated proportion of
extant species in state 0 and 1 that are included in the phylogeny.
A value of c(0.5, 0.75) means that half of species in state 0
and three quarters of species in state 1 are included in the
phylogeny. By default all species are assumed to be known.
The number of species modelled in unresolved clades (this is in addition to the largest observed clade).
List of control parameters for the ODE solver. See details below.
The states vector is always checked to make sure
that the values are 0 and 1 only. If strict is TRUE
(the default), then the additional check is made that every
state is present. The likelihood models tend to be poorly behaved
where states are missing.
Ratio of diversification rate to character change rate. Eventually this will be changed to allow for Mk2 to be used for estimating q parameters.
Logical: should starting parameters be Yule estimates rather than birth-death estimates?
Since 0.10.10 this is no longer supported. See the package README for more information.
This must be a data.frame with at least the four columns
tip.label, giving the name of the tip to which the data
applies
Nc, giving the number of species in the clade
n0, n1, giving the number of species known to be
in state 0 and 1, respectively.
These columns may be in any order, and additional columns will be ignored. (Note that column names are case sensitive).
An alternative way of specifying unresolved clade information is to
use the function make.clade.tree to construct a tree
where tips that represent clades contain information about which
species are contained within the clades. With a clade.tree,
the unresolved object will be automatically constructed from
the state information in states. (In this case, states
must contain state information for the species contained within the
unresolved clades.)
The differential equations that define the
BiSSE model are solved numerically using ODE solvers from the GSL
library or deSolve's LSODA. The control argument to
make.bisse controls the behaviour of the integrator. This is a
list that may contain elements:
tol: Numerical tolerance used for the calculations.
The default value of 1e-8 should be a reasonable trade-off
between speed and accuracy. Do not expect too much more than this
from the abilities of most machines!
eps: A value that when the sum of the D values drops
below, the integration results will be discarded and the
integration will be attempted again (the second-chance integration
will divide a branch in two and try again, recursively until the
desired accuracy is reached). The default value of 0 will
only discard integration results when the parameters go negative.
However, for some problems more restrictive values (on the order
of control$tol) will give better stability.
backend: Select the solver. The three options here are
gslode: (the default). Use the GSL solvers, by
default a Runge Kutta Kash Carp stepper.
deSolve: Use the LSODA solver from the
deSolve package. This is quite a bit slower at the
moment.
deSolve is the only supported backend on Windows.
Richard G. FitzJohn
make.bisse returns a function of class bisse. This
function has argument list (and default values)
f(pars, condition.surv=TRUE, root=ROOT.OBS, root.p=NULL,
intermediates=FALSE)
The arguments are interpreted as
pars A vector of six parameters, in the order
lambda0, lambda1, mu0, mu1,
q01, q10.
condition.surv (logical): should the likelihood
calculation condition on survival of two lineages and the speciation
event subtending them? This is done by default, following Nee et
al. 1994.
root: Behaviour at the root (see Maddison et al. 2007,
FitzJohn et al. 2009). The possible options are
ROOT.FLAT: A flat prior, weighting
\(D_0\) and \(D_1\) equally.
ROOT.EQUI: Use the equilibrium distribution
of the model, as described in Maddison et al. (2007).
ROOT.OBS: Weight \(D_0\) and
\(D_1\) by their relative probability of observing the
data, following FitzJohn et al. 2009:
$$D = D_0\frac{D_0}{D_0 + D_1} + D_1\frac{D_1}{D_0 + D_1}$$
ROOT.GIVEN: Root will be in state 0
with probability root.p[1], and in state 1 with
probability root.p[2].
ROOT.BOTH: Don't do anything at the root,
and return both values. (Note that this will not give you a
likelihood!).
root.p: Root weightings for use when
root=ROOT.GIVEN. sum(root.p) should equal 1.
intermediates: Add intermediates to the returned value as
attributes:
cache: Cached tree traversal information.
intermediates: Mostly branch end information.
vals: Root \(D\) values.
At this point, you will have to poke about in the source for more information on these.
starting.point.bisse produces a heuristic starting point to
start from, based on the character-independent birth-death model. You
can probably do better than this; see the vignette, for example.
bisse.starting.point is the same code, but deprecated in favour
of starting.point.bisse - it will be removed in a future
version.
FitzJohn R.G., Maddison W.P., and Otto S.P. 2009. Estimating trait-dependent speciation and extinction rates from incompletely resolved phylogenies. Syst. Biol. 58:595-611.
Maddison W.P., Midford P.E., and Otto S.P. 2007. Estimating a binary character's effect on speciation and extinction. Syst. Biol. 56:701-710.
Nee S., May R.M., and Harvey P.H. 1994. The reconstructed evolutionary process. Philos. Trans. R. Soc. Lond. B Biol. Sci. 344:305-311.
constrain for making submodels, find.mle
for ML parameter estimation, mcmc for MCMC integration,
and make.bd for state-independent birth-death models.
The help pages for find.mle has further examples of ML
searches on full and constrained BiSSE models.
## Due to a change in sample() behaviour in newer R it is necessary to
## use an older algorithm to replicate the previous examples
if (getRversion() >= "3.6.0") {
RNGkind(sample.kind = "Rounding")
}
pars <- c(0.1, 0.2, 0.03, 0.03, 0.01, 0.01)
set.seed(4)
phy <- tree.bisse(pars, max.t=30, x0=0)
## Here is the 52 species tree with the true character history coded.
## Red is state '1', which has twice the speciation rate of black (state
## '0').
h <- history.from.sim.discrete(phy, 0:1)
plot(h, phy)
lik <- make.bisse(phy, phy$tip.state)
lik(pars) # -159.71
Run the code above in your browser using DataLab