TreePar (version 3.3)

bd.shifts.optim: bd.shifts.optim: Estimating speciation and extinction rate changes and mass extinction events in phylogenies.

Description

bd.shifts.optim estimates the maximum likelihood speciation and extinction rates together with the rate shift times t=(t[1],t[2] .., t[m]) in a (possibly incomplete sampled) phylogeny. At the times t, the rates are allowed to change and the species may undergo a mass extinction event.

Usage

bd.shifts.optim(x, sampling, grid, start, end, maxitk = 5, yule = FALSE, ME = FALSE, 
all = FALSE, posdiv = FALSE, miniall = c(0), survival = 1,groups=0)

Arguments

x

Vector of speciation times in the phylogeny. Time is measured increasing going into the past with the present being time 0. x can be obtained from a phylogenetic tree using getx(TREE).

sampling

Vector of length m. sampling[i] is the probability of a species surviving the mass extinction at time t[i]. sampling[1] is the probability of an extant species being sampled. sampling[1]=1 means that the considered phylogeny is complete. sampling[i]=1 (i>1) means that at time t[i], a rate shift may occur but no species go extinct. If ME=TRUE, all entries but sampling[1] will be discarded as they are estimated. However, you have to input a vector sampling of the appropriate length such that the program knows how many mass extinction events you want to allow for.

grid, start, end

The model parameters are optimized for different fixed rate shift times. The fixed rate shift times are specified by being at (start, start+grid, start+2*grid .. end). I calculate the likelihood for the different rate shift times t instead of optimizing t with the function optim used for the other parameters, as the optimization performed poor for t (namely getting stuck in local optima).

yule

yule=TRUE sets the extinction rates to zero.

maxitk

Integer value defining how many iterations shall be done in the optimization. Default is 5, but needs to be increased if too many warnings "convergence problem" appear.

ME

ME=FALSE (default) uses the mass extinction fractions specified in sampling and does not estimate them. If ME=FALSE is used with sampling=c(1,1, .. , 1), no mass extinction events are considered.

all

Only relevant when ME=TRUE. all=FALSE (default and recommended) estimates one speciation and one extinction rate for the whole tree, and estimates the intensities sampling[i] (i>1) of mass extinction events. all=TRUE allows for varying speciation and extinction rates. Since the parameters might correlate, all=TRUE is not recommended.

posdiv

posdiv=FALSE (default) allows the (speciation - extinction) rate to be negative, i.e. allows for periods of declining diversity. posdiv=TRUE forces the (speciation - extinction) rate to be positive.

miniall

If you have run the bd.shifts.optim for k shifts, but you now want to have K>k shifts, then set for the subsequent analysis the following: update sampling, and set miniall=res[[2]] where res[[2]] is the output from the run with k shifts.

survival

If survival = 1, the likelihood is conditioned on survival of the process (recommended). Otherwise survival = 0.

groups

If groups != 0, the first column of groups indicates the age of higher taxa and the second column the number of species in the higher taxa (each row in groups corresponds to a leaf in the tree).

Value

res[[1]][[i]]

List of maximum likelihood parameter estimates for each fixed t (t determined by start, end, grid) where i-1 shifts are allowed to occur (i in 1:m). The first i entries are the turnover (extinction/speciation) estimates, for the successive intervals going back in time. The next i entries are the diversification rate estimates (speciation-extinction). The next i-1 entries are the probabilities that the lineage survives the mass extinction event (if ME=TRUE). (Note: if ME=TRUE and all=FALSE, the first entry is the turnover, the second the diversification rate, followed by the mass extinction survival probability).

res[[2]][[i]]

Maximum likelihood parameter estimates for i-1 shifts (i in 1:m). First entry is the (-log likelihood) value. The next entries are the maximum likelihood parameter estimates (see res[[1]][[i]]). The last i-1 entries are the shift times.

res[[3]]

Vector of time points where the function was evaluated.

res[[4]]

Array specifying the time points when there was a convergence problem: a row of res[[4]] with entry (i,t[i]) means that when adding the i-th shift at time t[i], a convergence problem was encountered.

References

T. Stadler. Mammalian phylogeny reveals recent diversification rate shifts. Proc. Nat. Acad. Sci., 108(15): 6187-6192, 2011.

T. Stadler, F. Bokma. Estimating speciation and extinction rates for phylogenies of higher taxa. Syst. Biol., 62(2): 220-230, 2013. (for groups>0 but no rate shift).

A. Lambert, T. Stadler. Macro-evolutionary models and coalescent point processes: the shape and probability of reconstructed phylogenies. Theo. Pop. Biol., 90: 113-128, 2013. (for groups>0).

Examples

set.seed(1)

# First we simulate a tree, and then estimate the parameters for the tree:
# Number of species
nspecies <- 20
# At time 1 and 2 in the past, we have a rate shift:
time <- c(0,1,2)
# Mass extinction intensities 0.5 at time 1 in past, 0.4 at time 2 in past. 
# Present day species are all sampled (rho[1]=1):
rho <- c(1,0.5,0.4)
# speciation rates (between t[i],t[i+1] we have speciation rate lambda[i]):
lambda <- c(2,2,1)
# extinction rates (between t[i],t[i+1] we have extinction rate mu[i]):
mu <- c(1,1,0)
# Simulation of a tree:
tree<-sim.rateshift.taxa(nspecies,1,lambda,mu,frac=rho,times=time,complete=FALSE)
# Extracting the speciation times x:
x<-sort(getx(tree[[1]]),decreasing=TRUE)

# When estimating the the rate shift times t based on branching times x, 
# we allow the shift times to be 0.6, 0.8, 1, 1.2, .. ,2.4:
start <- 0.6
end <- 2.4
grid <- 0.2


# We fix rho and estimate time, lambda, mu:
res <- bd.shifts.optim(x,rho,grid,start,end)[[2]]
res
# res[[2]] tells us about the maximum likelihood estimate given one rate shift:
# - log lik = 17.330862988.
# rate shift at time 2.2.
# turnover (extinction/speciation) = 0.186301549 more recent than 2.2,
#     and = 0.939681843 more ancestral than 2.2.
# net diversification (speciation-extinction) rate = 0.958947381 more recent than 2.2, 
#     and = 0.000100009 more ancestral than 2.2.

#test if i shifts explain the tree significantly better than i-1 shifts, here i=1:
i<-1
test<-pchisq(2*(res[[i]][1]-res[[i+1]][1]),3)
#if test>0.95 then i shifts is significantly better than i-1 shifts at a 5% error

# We fix rho=1 and mu=0 and then estimate time, lambda:
resyule <- bd.shifts.optim(x,rho,grid,start,end,yule=TRUE)
resyule[[2]]
# We estimate time, rho, lambda, mu:
resrho <- bd.shifts.optim(x,rho,grid,start,end,ME=TRUE)
resrho[[2]]
# Data analysis in Stadler & Bokma, 2012:
# Number of species in each order from Sibley and Monroe (1990)
data(bird.orders)
S <- c(10, 47, 69, 214, 161, 17, 355, 51, 56, 10, 39, 152, 6, 143, 
358, 103, 319, 23, 291, 313, 196, 1027, 5712)
groups<-get.groups(bird.orders,S,0)
groupscut<-get.groups(bird.orders,S,96.43)
x<-branching.times(bird.orders)
# transforming molecular timescale into calendar timescale
groups[,1]<-groups[,1]/0.207407
x<-x/0.207407 
bd.shifts.optim(x,sampling=c(1),survival=1,groups=groups)[[2]]
bd.shifts.optim(x,sampling=c(1),survival=1,groups=groupscut)[[2]]
# allowing one shift in rates:
bd.shifts.optim(x,sampling=c(1,1),grid=1,start=20,end=25,survival=1,groups=groupscut)[[2]]