fit_hbd_pdr_parametric {castor} R Documentation

## Fit parameterized pulled diversification rates of birth-death models.

### Description

Given an ultrametric timetree, estimate the pulled diversification rate (PDR) of homogenous birth-death (HBD) models that best explains the tree via maximum likelihood, assuming that the PDR is given as a parameterized function of time before present. Every HBD model is defined by some speciation and extinction rates (λ and μ) over time, as well as the sampling fraction ρ (fraction of extant species sampled). “Homogenous” refers to the assumption that, at any given moment in time, all lineages exhibit the same speciation/extinction rates. For any given HBD model there exists an infinite number of alternative HBD models that predict the same deterministic lineages-through-time curve and yield the same likelihood for any given reconstructed timetree; these “congruent” models cannot be distinguished from one another solely based on the tree.

Each congruence class is uniquely described by its PDR, defined as PDR=λ-μ+λ^{-1}dλ/dτ (where τ is time before present) as well as the product ρλ_o (where λ_o is the present-day speciation rate). That is, two HBD models are congruent if and only if they have the same PDR and the same product ρλ_o. This function is designed to estimate the generating congruence class for the tree, by fitting a finite number of parameters defining the PDR and ρλ_o.

### Usage

fit_hbd_pdr_parametric( tree,
param_values,
param_guess        = NULL,
param_min          = -Inf,
param_max          = +Inf,
param_scale        = NULL,
oldest_age         = NULL,
age0               = 0,
PDR,
rholambda0,
age_grid           = NULL,
condition          = "auto",
relative_dt        = 1e-3,
Ntrials            = 1,
max_start_attempts = 1,
max_model_runtime  = NULL,
fit_control        = list())


### Arguments

 tree An ultrametric timetree of class "phylo", representing the time-calibrated phylogeny of a set of extant species. param_values Numeric vector, specifying fixed values for a some or all model parameters. For fitted (i.e., non-fixed) parameters, use NaN or NA. For example, the vector c(1.5,NA,40) specifies that the 1st and 3rd model parameters are fixed at the values 1.5 and 40, respectively, while the 2nd parameter is to be fitted. The length of this vector defines the total number of model parameters. If entries in this vector are named, the names are taken as parameter names. Names should be included if you'd like returned parameter vectors to have named entries, or if the functions PDR or rho query parameter values by name (as opposed to numeric index). param_guess Numeric vector of size NP, specifying a first guess for the value of each model parameter. For fixed parameters, guess values are ignored. Can be NULL only if all model parameters are fixed. param_min Optional numeric vector of size NP, specifying lower bounds for model parameters. If of size 1, the same lower bound is applied to all parameters. Use -Inf to omit a lower bound for a parameter. If NULL, no lower bounds are applied. For fixed parameters, lower bounds are ignored. param_max Optional numeric vector of size NP, specifying upper bounds for model parameters. If of size 1, the same upper bound is applied to all parameters. Use +Inf to omit an upper bound for a parameter. If NULL, no upper bounds are applied. For fixed parameters, upper bounds are ignored. param_scale Optional numeric vector of size NP, specifying typical scales for model parameters. If of size 1, the same scale is assumed for all parameters. If NULL, scales are determined automatically. For fixed parameters, scales are ignored. It is strongly advised to provide reasonable scales, as this facilitates the numeric optimization algorithm. oldest_age Strictly positive numeric, specifying the oldest time before present (“age”) to consider when calculating the likelihood. If this is equal to or greater than the root age, then oldest_age is taken as the stem age, and the classical formula by Morlon et al. (2011) is used. If oldest_age is less than the root age, the tree is split into multiple subtrees at that age by treating every edge crossing that age as the stem of a subtree, and each subtree is considered an independent realization of the HBD model stemming at that age. This can be useful for avoiding points in the tree close to the root, where estimation uncertainty is generally higher. If oldest_age==NULL, it is automatically set to the root age. age0 Non-negative numeric, specifying the youngest age (time before present) to consider for fitting, and with respect to which rholambda0 is defined. If age0>0, then rholambda0 refers to the product of the sampling fraction at age age0 and the speciation rate at age age0. See below for more details. PDR Function specifying the pulled diversification rate at any given age (time before present) and for any given parameter values. This function must take exactly two arguments, the 1st one being a numeric vector (one or more ages) and the 2nd one being a numeric vector of size NP (parameter values), and return a numeric vector of the same size as the 1st argument. Can also be a single number (i.e., PDR is fixed). rholambda0 Function specifying the product ρλ_o (sampling fraction times speciation rate at age0) for any given parameter values. This function must take exactly one argument, a numeric vector of size NP (parameter values), and return a strictly positive numeric. Can also be a single number (i.e., rholambda0 is fixed). age_grid Numeric vector, specifying ages at which the PDR function should be evaluated. This age grid must be fine enough to capture the possible variation in the PDR over time, within the permissible parameter range. If of size 1, then the PDR is assumed to be time-independent. Listed ages must be strictly increasing, and must cover at least the full considered age interval (from age0 to oldest_age). Can also be NULL or a vector of size 1, in which case the PDR is assumed to be time-independent. condition Character, either "crown", "stem" or "auto", specifying on what to condition the likelihood. If "crown", the likelihood is conditioned on the survival of the two daughter lineages branching off at the root. If "stem", the likelihood is conditioned on the survival of the stem lineage. Note that "crown" really only makes sense when oldest_age is equal to the root age, while "stem" is recommended if oldest_age differs from the root age. If "auto", the condition is chosen according to the recommendations mentioned earlier. relative_dt Strictly positive numeric (unitless), specifying the maximum relative time step allowed for integration over time, when calculating the likelihood. Smaller values increase integration accuracy but increase computation time. Typical values are 0.0001-0.001. The default is usually sufficient. Ntrials Integer, specifying the number of independent fitting trials to perform, each starting from a random choice of model parameters. Increasing Ntrials reduces the risk of reaching a non-global local maximum in the fitting objective. max_start_attempts Integer, specifying the number of times to attempt finding a valid start point (per trial) before giving up on that trial. Randomly choosen extreme start parameters may occasionally result in Inf/undefined likelihoods, so this option allows the algorithm to keep looking for valid starting points. Nthreads Integer, specifying the number of parallel threads to use for performing multiple fitting trials simultaneously. This should generally not exceed the number of available CPUs on your machine. Parallel computing is not available on the Windows platform. max_model_runtime Optional numeric, specifying the maximum number of seconds to allow for each evaluation of the likelihood function. Use this to abort fitting trials leading to parameter regions where the likelihood takes a long time to evaluate (these are often unlikely parameter regions). fit_control Named list containing options for the nlminb optimization routine, such as iter.max, eval.max or rel.tol. For a complete list of options and default values see the documentation of nlminb in the stats package.

### Details

This function is designed to estimate a finite set of scalar parameters (p_1,..,p_n\in\R) that determine the PDR and the product ρλ_o (sampling fraction times present-dat extinction rate), by maximizing the likelihood of observing a given timetree under the HBD model. For example, the investigator may assume that the PDR varies exponentially over time, i.e. can be described by PDR(t)=A\cdot e^{-B t} (where A and B are unknown coefficients and t is time before present), and that the product ρλ_o is unknown. In this case the model has 3 free parameters, p_1=A, p_2=B and p_3=ρλ_o, each of which may be fitted to the tree.

If age0>0, the input tree is essentially trimmed at age0 (omitting anything younger than age0), and the PDR and rholambda0 are fitted to this new (shorter) tree, with time shifted appropriately. The fitted rholambda0 is thus the product of the sampling fraction at age0 and the speciation rate at age0. Note that the sampling fraction at age0 is simply the fraction of lineages extant at age0 that are represented in the timetree.

It is generally advised to provide as much information to the function fit_hbd_pdr_parametric as possible, including reasonable lower and upper bounds (param_min and param_max), a reasonable parameter guess (param_guess) and reasonable parameter scales param_scale. If some model parameters can vary over multiple orders of magnitude, it is advised to transform them so that they vary across fewer orders of magnitude (e.g., via log-transformation). It is also important that the age_grid is sufficiently fine to capture the variation of the PDR over time, since the likelihood is calculated under the assumption that both vary linearly between grid points.

### Value

A list with the following elements:

 success Logical, indicating whether model fitting succeeded. If FALSE, the returned list will include an additional “error” element (character) providing a description of the error; in that case all other return variables may be undefined. objective_value The maximized fitting objective. Currently, only maximum-likelihood estimation is implemented, and hence this will always be the maximized log-likelihood. objective_name The name of the objective that was maximized during fitting. Currently, only maximum-likelihood estimation is implemented, and hence this will always be “loglikelihood”. param_fitted Numeric vector of size NP (number of model parameters), listing all fitted or fixed model parameters in their standard order (see details above). If param_names was provided, elements in fitted_params will be named. param_guess Numeric vector of size NP, listing guessed or fixed values for all model parameters in their standard order. loglikelihood The log-likelihood of the fitted model for the given timetree. NFP Integer, number of fitted (i.e., non-fixed) model parameters. AIC The Akaike Information Criterion for the fitted model, defined as 2k-2\log(L), where k is the number of fitted parameters and L is the maximized likelihood. BIC The Bayesian information criterion for the fitted model, defined as \log(n)k-2\log(L), where k is the number of fitted parameters, n is the number of data points (number of branching times), and L is the maximized likelihood. converged Logical, specifying whether the maximum likelihood was reached after convergence of the optimization algorithm. Note that in some cases the maximum likelihood may have been achieved by an optimization path that did not yet converge (in which case it's advisable to increase iter.max and/or eval.max). Niterations Integer, specifying the number of iterations performed during the optimization path that yielded the maximum likelihood. Nevaluations Integer, specifying the number of likelihood evaluations performed during the optimization path that yielded the maximum likelihood. trial_start_objectives Numeric vector of size Ntrials, listing the initial objective values (e.g., loglikelihoods) for each fitting trial, i.e. at the start parameter values. trial_objective_values Numeric vector of size Ntrials, listing the final maximized objective values (e.g., loglikelihoods) for each fitting trial. trial_Nstart_attempts Integer vector of size Ntrials, listing the number of start attempts for each fitting trial, until a starting point with valid likelihood was found. trial_Niterations Integer vector of size Ntrials, listing the number of iterations needed for each fitting trial. trial_Nevaluations Integer vector of size Ntrials, listing the number of likelihood evaluations needed for each fitting trial.

Stilianos Louca

### References

H. Morlon, T. L. Parsons, J. B. Plotkin (2011). Reconciling molecular phylogenies with the fossil record. Proceedings of the National Academy of Sciences. 108:16327-16332.

S. Louca et al. (2018). Bacterial diversification through geological time. Nature Ecology & Evolution. 2:1458-1467.

simulate_deterministic_hbd

loglikelihood_hbd

fit_hbd_model_on_grid

fit_hbd_model_parametric

fit_hbd_pdr_on_grid

### Examples

## Not run:
# Generate a random tree with exponentially varying lambda & mu
Ntips     = 10000
rho       = 0.5 # sampling fraction
time_grid = seq(from=0, to=100, by=0.01)
lambdas   = 2*exp(0.1*time_grid)
mus       = 1.5*exp(0.09*time_grid)
tree      = generate_random_tree( parameters  = list(rarefaction=rho),
max_tips    = Ntips/rho,
coalescent  = TRUE,
added_death_rates_pc  = mus)$tree root_age = castor::get_tree_span(tree)$max_distance
cat(sprintf("Tree has %d tips, spans %g Myr\n",length(tree$tip.label),root_age)) # Define a parametric HBD congruence class, with exponentially varying PDR # The model thus has 3 parameters PDR_function = function(ages,params){ return(params['A']*exp(-params['B']*ages)); } rholambda0_function = function(params){ return(params['rholambda0']) } # Define an age grid on which lambda_function & mu_function shall be evaluated # Should be sufficiently fine to capture the variation in the PDR age_grid = seq(from=0,to=100,by=0.01) # Perform fitting # Lets suppose extinction rates are already known cat(sprintf("Fitting class to tree..\n")) fit = fit_hbd_pdr_parametric( tree, param_values = c(A=NA, B=NA, rholambda0=NA), param_guess = c(1,0,1), param_min = c(-10,-10,0), param_max = c(10,10,10), param_scale = 1, # all params are in the order of 1 PDR = PDR_function, rholambda0 = rholambda0_function, age_grid = age_grid, Ntrials = 10, # perform 10 fitting trials Nthreads = 2, # use 2 CPUs max_model_runtime = 1, # limit model evaluation to 1 second fit_control = list(rel.tol=1e-6)) if(!fit$success){
cat(sprintf("ERROR: Fitting failed: %s\n",fit$error)) }else{ cat(sprintf("Fitting succeeded:\nLoglikelihood=%g\n",fit$loglikelihood))