| Title: | Coalescent-Based Simulation of Ecological Communities |
|---|---|
| Description: | Coalescent-Based Simulation of Ecological Communities as proposed by Munoz et al. (2018) <doi:10.1111/2041-210X.12918>. The package includes a tool for estimating parameters of community assembly by using Approximate Bayesian Computation. |
| Authors: | François Munoz [aut, cre], Matthias Grenié [aut], Pierre Denelle [aut], Elizabeth Barthelemy [aut], Adrien Taudière [ctb], Fabien Laroche [ctb], Caroline Tucker [ctb], Cyrille Violle [ctb] |
| Maintainer: | François Munoz <[email protected]> |
| License: | GPL-3 |
| Version: | 2.0.0 |
| Built: | 2024-10-26 03:12:26 UTC |
| Source: | https://github.com/frmunoz/ecolottery |
Coalescent-Based Simulation of Ecological Communities as proposed by Munoz et al. (2018) <doi:10.1111/2041-210X.12918>. The package includes a tool for estimating parameters of community assembly by using Approximate Bayesian Computation.
The DESCRIPTION file:
| Package: | ecolottery |
| Type: | Package |
| Title: | Coalescent-Based Simulation of Ecological Communities |
| Version: | 2.0.0 |
| Authors@R: | c(person("François", "Munoz", role = c("aut", "cre"), email = "[email protected]"), person("Matthias", "Grenié", role = "aut"), person("Pierre", "Denelle", role = "aut"), person("Elizabeth", "Barthelemy", role = "aut"), person("Adrien", "Taudière", role = "ctb"), person("Fabien", "Laroche", role = "ctb"), person("Caroline", "Tucker", role = "ctb"), person("Cyrille", "Violle", role = "ctb")) |
| URL: | https://github.com/frmunoz/ecolottery |
| BugReports: | https://github.com/frmunoz/ecolottery/issues |
| BuildVignettes: | true |
| VignetteBuilder: | knitr |
| Depends: | R (>= 3.0.2) |
| Imports: | abc, ade4, data.table, plyr, stats, graphics, ggplot2, grDevices, hillR, parallel, EasyABC, lazyeval, moments, pbapply, Rcpp |
| Suggests: | ape, dplyr, e1071, changepoint, knitr, picante, RColorBrewer, reshape, rmarkdown, sads, testthat, utils, vegan, Weighted.Desc.Stat |
| #LinkingTo: | Rcpp |
| Description: | Coalescent-Based Simulation of Ecological Communities as proposed by Munoz et al. (2018) <doi:10.1111/2041-210X.12918>. The package includes a tool for estimating parameters of community assembly by using Approximate Bayesian Computation. |
| License: | GPL-3 |
| Encoding: | UTF-8 |
| RoxygenNote: | 6.1.1 |
| Repository: | https://rekyt.r-universe.dev |
| RemoteUrl: | https://github.com/frmunoz/ecolottery |
| RemoteRef: | HEAD |
| RemoteSha: | cbd4fe296d00768fe8a523430a26a359e7d41fea |
| Author: | François Munoz [aut, cre], Matthias Grenié [aut], Pierre Denelle [aut], Elizabeth Barthelemy [aut], Adrien Taudière [ctb], Fabien Laroche [ctb], Caroline Tucker [ctb], Cyrille Violle [ctb] |
| Maintainer: | François Munoz <[email protected]> |
Index of help topics:
abund Compute absolute and relative abundances in the
local community and the reference pool
coalesc Coalescent-based simulation of ecological
communities undergoing both neutral and
niche-base dynamics
coalesc_abc Estimation of parameters of community assembly
using Approximate Bayesian Computation (ABC)
coalesc_abc_std Estimation of parameters of community assembly
using Approximate Bayesian Computation (ABC)
with random exploration of prior parameter
distributions
coalesc_easyABC Estimation of parameters of community assembly
using Approximate Bayesian Computation (ABC),
with algorithms of optimal exploration of prior
distributions
ecolottery-package Coalescent-Based Simulation of Ecological
Communities
forward Simulation of neutral and niche-based community
dynamics forward in time
plot_comm Plotting trait and species distributions of
simulated communities
prdch4coalesc Posterior predictive checks using
coalescent-based simulation of ecological
communities
sumstats Predifined summary statistic function for
estimating neutral and non-neutral parameters
of community assembly using Approximate
Bayesian Computation (ABC) implemented in
coalesc_abc2.
tcor Generates Correlated Traits
Two basic functions: coalesc for coalescent-based simulation, and forward for forward-in-time simulation
François Munoz [aut, cre], Matthias Grenié [aut], Pierre Denelle [aut], Elizabeth Barthelemy [aut], Adrien Taudière [ctb], Fabien Laroche [ctb], Caroline Tucker [ctb], Cyrille Violle [ctb]
Maintainer: François Munoz <[email protected]>
Hurtt, G. C. and S. W. Pacala (1995). "The consequences of recruitment limitation: reconciling chance, history and competitive differences between plants." Journal of Theoretical Biology 176(1): 1-12.
Hubbell, S. P. (2001). "The Unified Neutral Theory of Biodiversity". Princeton University Press.
Gravel, D., C. D. Canham, M. Beaudet and C. Messier (2006). "Reconciling niche and neutrality: the continuum hypothesis." Ecology Letters 9(4): 399-409.
Munoz, F., P. Couteron, B. R. Ramesh and R. S. Etienne (2007). "Estimating parameters of neutral communities: from one Single Large to Several Small samples." Ecology 88(10): 2482-2488.
Munoz, F., B. R. Ramesh and P. Couteron (2014). "How do habitat filtering and niche conservatism affect community composition at different taxonomic resolutions?" Ecology 95(8): 2179-2191.
## Coalescent-based simulation of stabilizing habitat filtering around ## t = 0.5 J <- 100; theta <- 50; m <- 0.5; comm <- coalesc(J, m, theta, filt = function(x) 0.5 - abs(0.5 - x)) plot_comm(comm) ## Forward-in-time simulation of stabilizing habitat filtering around ## t = 0.5, over 100 time steps # A regional pool including 100 species each including 10 individuals pool <- sort(rep(as.character(1:100), 10)) # Initial community composed of 10 species each including 10 individuals, # with trait information for niche-based dynamics initial <- data.frame(ind = paste("init", 1:100, sep="."), sp = sort(rep(as.character(1:10), 10)), trait = runif(100)) final <- forward(initial = initial, m = 0.5, gens = 100, pool = pool, filt = function(x) 0.5 - abs(0.5 - x)) plot_comm(final)## Coalescent-based simulation of stabilizing habitat filtering around ## t = 0.5 J <- 100; theta <- 50; m <- 0.5; comm <- coalesc(J, m, theta, filt = function(x) 0.5 - abs(0.5 - x)) plot_comm(comm) ## Forward-in-time simulation of stabilizing habitat filtering around ## t = 0.5, over 100 time steps # A regional pool including 100 species each including 10 individuals pool <- sort(rep(as.character(1:100), 10)) # Initial community composed of 10 species each including 10 individuals, # with trait information for niche-based dynamics initial <- data.frame(ind = paste("init", 1:100, sep="."), sp = sort(rep(as.character(1:10), 10)), trait = runif(100)) final <- forward(initial = initial, m = 0.5, gens = 100, pool = pool, filt = function(x) 0.5 - abs(0.5 - x)) plot_comm(final)
Compute the abundances and relative abundances of species in simulated
communities and in the corresponding species pools. The input must be an output of
either coalesc or the forward functions.
abund(x)abund(x)
x |
a list including the species pool composition ( |
pool |
species abundances and relative abundances in the reference pool |
com |
species abundances and relative abundances in the local community |
F. Munoz, P. Denelle and M. Grenie
# Simulation of a neutral community including 500 individuals J <- 500; theta <- 50; m <- 0.05; comm1a <- coalesc(J, m, theta) abund1a <- abund(comm1a) # Log-series distribution of regional abundances fit <- vegan::fisherfit(abund1a$pool$ab) freq <- as.numeric(names(fit$fisher)) plot(log(freq), fit$fisher, xlab = "Frequency (log)", ylab = "Species", type = "n") rect(log(freq - 0.5), 0, log(freq + 0.5), fit$fisher, col="skyblue") alpha <- fit$estimate k <- fit$nuisance curve(alpha * k^exp(x) / exp(x), log(0.5), max(log(freq)), col = "red", lwd = 2, add = TRUE) # Relationship between local and regional abundances par(mfrow=c(1, 2)) plot(abund1a$pool[rownames(abund1a$com), "relab"], abund1a$com$relab, main = "m = 0.05", xlab = "Regional abundance", ylab = "Local abundance", log = "xy") abline(0,1) # With higher immigration rate m <- 0.95 comm1b <- coalesc(J, m, theta) abund1b <- abund(comm1b) plot(abund1b$pool[rownames(abund1b$com),"relab"], abund1b$com$relab, main = "m = 0.95", xlab = "Regional abundance", ylab = "Local abundance", log = "xy") abline(0,1)# Simulation of a neutral community including 500 individuals J <- 500; theta <- 50; m <- 0.05; comm1a <- coalesc(J, m, theta) abund1a <- abund(comm1a) # Log-series distribution of regional abundances fit <- vegan::fisherfit(abund1a$pool$ab) freq <- as.numeric(names(fit$fisher)) plot(log(freq), fit$fisher, xlab = "Frequency (log)", ylab = "Species", type = "n") rect(log(freq - 0.5), 0, log(freq + 0.5), fit$fisher, col="skyblue") alpha <- fit$estimate k <- fit$nuisance curve(alpha * k^exp(x) / exp(x), log(0.5), max(log(freq)), col = "red", lwd = 2, add = TRUE) # Relationship between local and regional abundances par(mfrow=c(1, 2)) plot(abund1a$pool[rownames(abund1a$com), "relab"], abund1a$com$relab, main = "m = 0.05", xlab = "Regional abundance", ylab = "Local abundance", log = "xy") abline(0,1) # With higher immigration rate m <- 0.95 comm1b <- coalesc(J, m, theta) abund1b <- abund(comm1b) plot(abund1b$pool[rownames(abund1b$com),"relab"], abund1b$com$relab, main = "m = 0.95", xlab = "Regional abundance", ylab = "Local abundance", log = "xy") abline(0,1)
Simulates the composition of a community based on immigration from a regional pool, habitat filtering depending on local environment and species traits, and local birth-death stochastic dynamics.
coalesc(J, m = 1, theta = NULL, filt = NULL, filt.vect = F, m.replace = T, add = FALSE, var.add =NULL, pool = NULL, traits = NULL, Jpool = 50 * J, verbose = FALSE, checks = TRUE)coalesc(J, m = 1, theta = NULL, filt = NULL, filt.vect = F, m.replace = T, add = FALSE, var.add =NULL, pool = NULL, traits = NULL, Jpool = 50 * J, verbose = FALSE, checks = TRUE)
J |
number of individuals in the local community. |
m |
migration rate (if |
theta |
parameter of neutral dynamics in the regional pool (used only if |
m.replace |
should the immigrants be drawn with replacement from the source pool. Default is |
filt |
a function representing the effect of local habitat filtering. For an individual that displays the value(s) |
filt.vect |
indicates whether the filtering function can be vectorized. It means that the function can take as input a vector of trait values and provide a vector of the corresponding weights. |
add |
indicates if additional variables must be passed to |
var.add |
additional variables to be passed to |
pool |
the regional pool of species providing immigrants to the local community. It should include the label of individual on first column, and of its species on second column. If |
traits |
a matrix or data.frame including one or several traits on columns. A unique trait value is assigned to each species in the regional pool. Species names of |
Jpool |
if |
verbose |
if |
checks |
should initial checks that the inputs are correct be performed? |
Coalescent-based simulation of a community of size J. This generic function can simulate a neutral community (if filt = NULL) or a community undergoing both neutral and niche-based dynamics. In the latter case, filt(t) represents the relative ability of immigrants with trait values t in the regional pool to enter the community.
If trait-based filtering involves several traits, filt(t) must be defined such as t is a vector of the trait values of a candidate immigrant.
See example below.
A Shiny app is available to visualize simulated trait distributions for chosen parameter values in the model.
com |
a data.frame of simulated individuals, with the label of ancestor individual in the regional pool on first column (as in first column of pool), species label on second column (as in second column of pool), and species trait (as in third column of pool).
Not provided if |
pool |
a data.frame of the individuals of the regional source pool, with the label of ancestor individual in the regional pool on first column (as in first column of input |
call |
the call function. |
F. Munoz
Hurtt, G. C. and S. W. Pacala (1995). "The consequences of recruitment limitation: reconciling chance, history and competitive differences between plants." Journal of Theoretical Biology 176(1): 1-12.
Gravel, D., C. D. Canham, M. Beaudet and C. Messier (2006). "Reconciling niche and neutrality: the continuum hypothesis." Ecology Letters 9(4): 399-409.
Munoz, F., P. Couteron, B. R. Ramesh and R. S. Etienne (2007). "Estimating parameters of neutral communities: from one Single Large to Several Small samples." Ecology 88(10): 2482-2488.
Munoz, F., B. R. Ramesh and P. Couteron (2014). "How do habitat filtering and niche conservatism affect community composition at different taxonomic resolutions?" Ecology 95(8): 2179-2191.
# Simulation of a neutral community including 100 individuals J <- 500; theta <- 50; m <- 0.1 comm1 <- coalesc(J, m, theta) # Regional and local trait distributions plot_comm(comm1) # Define a regional pool of species with equal abundances pool <- cbind(1:10000, rep(1:500, 20), rep(NA, 10000)) # Uniform distribution of trait values t.sp <- runif(500) # No intraspecific variation pool[,3] <- t.sp[pool[,2]] # Generate a neutral community drawn from the pool comm2<- coalesc(J, m, pool = pool) plot_comm(comm2) # Directional habitat filtering toward t = 0 comm3 <- coalesc(J, m, filt = function(x) 1 - x, pool = pool) # Regional and local trait distributions plot_comm(comm3) # Function for environmental filtering sigma <- 0.1 filt_gaussian <- function(t, x) exp(-(x - t)^2/(2*sigma^2)) # Stabilizing habitat filtering around t = 0.1 comm4a <- coalesc(J, m, filt = function(x) filt_gaussian(0.1, x), pool = pool) plot_comm(comm4a) # Stabilizing habitat filtering around t = 0.5 comm4b <- coalesc(J, m, filt = function(x) filt_gaussian(0.5, x), pool = pool) plot_comm(comm4b) # Stabilizing habitat filtering around t = 0.9 comm4c <- coalesc(J, m, filt = function(x) filt_gaussian(0.9, x), pool = pool) plot_comm(comm4c) # Mean trait values in communities reflect the influence of habitat filtering mean(comm4a$com[, 3]) mean(comm4b$com[, 3]) mean(comm4c$com[, 3]) # Disruptive habitat filtering around t = 0.5 comm5 <- coalesc(J, m, filt = function(x) abs(0.5 - x), pool = pool) plot_comm(comm5) # Multi-modal habitat filtering t.sp <- rnorm(500) pool[, 3] <- t.sp[pool[,2]] comm6 <- coalesc(J, m, filt = function(x) sin(3*x) + 1, pool = pool) plot_comm(comm6) # Filtering depending on multiple traits # We define two traits t1.sp <- runif(500) t2.sp <- runif(500) pool[, 3] <- t1.sp[pool[,2]] pool <- cbind(pool, t2.sp[pool[,2]]) # Here the probability of successful immigration depends # on the product of Gaussian filters playing on two traits, # with different optimal values comm7 <- coalesc(J, m, filt = function(x) filt_gaussian(0.75, x[1])*filt_gaussian(0.25, x[2]), pool = pool) # Distribution of trait 1 plot_comm(comm7, seltrait = 1) # Distribution of trait 2 plot_comm(comm7, seltrait = 2)# Simulation of a neutral community including 100 individuals J <- 500; theta <- 50; m <- 0.1 comm1 <- coalesc(J, m, theta) # Regional and local trait distributions plot_comm(comm1) # Define a regional pool of species with equal abundances pool <- cbind(1:10000, rep(1:500, 20), rep(NA, 10000)) # Uniform distribution of trait values t.sp <- runif(500) # No intraspecific variation pool[,3] <- t.sp[pool[,2]] # Generate a neutral community drawn from the pool comm2<- coalesc(J, m, pool = pool) plot_comm(comm2) # Directional habitat filtering toward t = 0 comm3 <- coalesc(J, m, filt = function(x) 1 - x, pool = pool) # Regional and local trait distributions plot_comm(comm3) # Function for environmental filtering sigma <- 0.1 filt_gaussian <- function(t, x) exp(-(x - t)^2/(2*sigma^2)) # Stabilizing habitat filtering around t = 0.1 comm4a <- coalesc(J, m, filt = function(x) filt_gaussian(0.1, x), pool = pool) plot_comm(comm4a) # Stabilizing habitat filtering around t = 0.5 comm4b <- coalesc(J, m, filt = function(x) filt_gaussian(0.5, x), pool = pool) plot_comm(comm4b) # Stabilizing habitat filtering around t = 0.9 comm4c <- coalesc(J, m, filt = function(x) filt_gaussian(0.9, x), pool = pool) plot_comm(comm4c) # Mean trait values in communities reflect the influence of habitat filtering mean(comm4a$com[, 3]) mean(comm4b$com[, 3]) mean(comm4c$com[, 3]) # Disruptive habitat filtering around t = 0.5 comm5 <- coalesc(J, m, filt = function(x) abs(0.5 - x), pool = pool) plot_comm(comm5) # Multi-modal habitat filtering t.sp <- rnorm(500) pool[, 3] <- t.sp[pool[,2]] comm6 <- coalesc(J, m, filt = function(x) sin(3*x) + 1, pool = pool) plot_comm(comm6) # Filtering depending on multiple traits # We define two traits t1.sp <- runif(500) t2.sp <- runif(500) pool[, 3] <- t1.sp[pool[,2]] pool <- cbind(pool, t2.sp[pool[,2]]) # Here the probability of successful immigration depends # on the product of Gaussian filters playing on two traits, # with different optimal values comm7 <- coalesc(J, m, filt = function(x) filt_gaussian(0.75, x[1])*filt_gaussian(0.25, x[2]), pool = pool) # Distribution of trait 1 plot_comm(comm7, seltrait = 1) # Distribution of trait 2 plot_comm(comm7, seltrait = 2)
Estimates parameters of neutral migration-drift dynamics (through migration
rate m and parameters of environmental filtering (through a filtering function
filt.abc()) from the composition of a local community and the related
regional pool.
coalesc_abc(comm.obs, pool = NULL, multi = "single", prop = FALSE, traits = NULL, f.sumstats, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = TRUE, size.abc = NULL, add = FALSE, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = FALSE, dim.pca = NULL, svd = FALSE, theta.max = NULL, nb.samp = 10^6, parallel = FALSE, nb.core = NULL, tol = NULL, type = "standard", method.seq = "Lenormand", method.mcmc = "Marjoram_original", method.abc = "rejection", alpha = 0.5, pkg = NULL) initial_checks(comm.obs = NULL, pool = NULL, multi = "single", prop = F, traits = NULL, f.sumstats = NULL, filt.abc = NULL, filt.vect = F, migr.abc = NULL, size.abc = NULL, add = NULL, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = F, dim.pca = NULL, svd = NULL, theta.max = NULL, nb.samp = 10^6, parallel = F, nb.core = NULL, tol = NULL, type = "standard", method.seq = "Lenormand", method.mcmc = "Marjoram_original", method.abc = "rejection", alpha = 0.5, pkg = NULL)coalesc_abc(comm.obs, pool = NULL, multi = "single", prop = FALSE, traits = NULL, f.sumstats, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = TRUE, size.abc = NULL, add = FALSE, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = FALSE, dim.pca = NULL, svd = FALSE, theta.max = NULL, nb.samp = 10^6, parallel = FALSE, nb.core = NULL, tol = NULL, type = "standard", method.seq = "Lenormand", method.mcmc = "Marjoram_original", method.abc = "rejection", alpha = 0.5, pkg = NULL) initial_checks(comm.obs = NULL, pool = NULL, multi = "single", prop = F, traits = NULL, f.sumstats = NULL, filt.abc = NULL, filt.vect = F, migr.abc = NULL, size.abc = NULL, add = NULL, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = F, dim.pca = NULL, svd = NULL, theta.max = NULL, nb.samp = 10^6, parallel = F, nb.core = NULL, tol = NULL, type = "standard", method.seq = "Lenormand", method.mcmc = "Marjoram_original", method.abc = "rejection", alpha = 0.5, pkg = NULL)
comm.obs |
the observed community composition. If |
pool |
composition of the regional pool to which the local community is
hypothesized to be related through migration dynamics with possible
environmental filtering. Should be a matrix of individuals on rows with
their individual id (first column), species id (second column), and
(optionally) the trait values of the individuals.
When |
multi |
structure of the community inputs:
|
prop |
indicates if the community composition is given in term of relative species abundances,
cover or proportions. In this case, a parameter of effective community size is estimated.
Default is FALSE. If no prior is provided for |
traits |
the trait values of species in the regional pool. It is used if trait
information is not provided in |
f.sumstats |
a function calculating the summary statistics of community
composition. It is be used to compare observed and simulated community
composition in ABC estimation. The first input argument of the function concerns community
composition, under a format depending on |
filt.abc |
the hypothesized environmental filtering function. It is a function of
individual trait values (first argument), additional parameters to be estimated and (optionally)
environmental variables defined in |
filt.vect |
indicates whether the filtering function can be vectorized. It means that the function can take as input a vector of trait values and provide a vector of the corresponding weights. |
migr.abc |
a function defining immigration probability in local communities. It can be a function of
environmental variables defined in |
m.replace |
should the immigrants be drawn with replacement from the source pool. Default is |
size.abc |
a function defining local community sizes. It can be used when |
add |
indicates if additional variables must be passed to |
var.add |
additional variables to be passed to |
params |
equivalent to par.filt (see below): it is kept in the function for compatibility with previous versions. |
par.filt |
a matrix of the bounds of the parameters used in |
par.migr |
a matrix of the bounds of the parameters used in |
par.size |
a matrix of the bounds of the parameters used in |
constr |
constraints that must be set on parameter values, i.e., relationships that must
be met when drawing the values in prior distributions. It must be a vector of
character strings each including an operator relating some of the parameters.
The names of parameters must be consistent with those used in |
scale |
should the summary statistics be scaled. Default is |
dim.pca |
gives a number of independent dimensions to calculate from the simulated summary statistics,
by performing Principal Component Analysis (PCA). Default is |
svd |
indicates if Singular Value Decomposition (SVD) must be performed to get the basic
dimensions requested in |
theta.max |
if |
nb.samp |
the number of parameter values to be sampled in ABC calculation. Random
values of parameters of environmental filtering (see |
parallel |
boolean. If |
nb.core |
number of cores to be used in parallel computation if |
tol |
the tolerance value used in ABC estimation (see help in
|
pkg |
packages needed for calculation of |
type |
the type of algorithm to be used in EasyABC. Can be either "standard" (using package abc, default)" "seq" (sequential), "mcmc" or "annealing". Three later options are based on the EasyABC package. |
method.seq |
when type = "seq", gives the algorithm for sequential sampling scheme, which is passed to |
method.mcmc |
when type = "mcmc", gives the algorithm for MCMC sampling scheme, which is passed to |
method.abc |
the method to be used in ABC estimation (see help on
|
alpha |
a positive number between 0 and 1 (strictly) used when performing sequential ABC method. |
coalesc_abc() performs ABC estimation for one (if multi = "single",
default) or several communities (if multi = "tab" or "seqcom") related to the same
regional pool. If type = "standard", it performs simulations with parameter values randomly drawn in priori distributions. With other options in type, the function uses optimization algorithms implemented in the package EasyABC.
To assess environmental filtering, a filtering function must be defined in filt.abc. It should take two arguments, the first is a trait value of a candidate immigrant, the second is a vector including the parameter values of the filtering function. See examples below for further information on usage.
An important point is that in many cases the function might be vectorized, that is, it can provide a vector of filtering probabilities for a vector of trait values given in first argument. In this case the user should set filt.vect = T, which will significantly accelerate simulations.
The related functions coalesc_abc_std() and coalesc_abc_std() perform simulations with the option type = "standard" and with the algorithms from EasyABC package, respectivelt.
initial_checks() performs initial checks of the input arguments of coalesc_abc. It provides informative error and warning messages when needed.
This function is not intended to be used directly.
par |
parameter values used in simulations. |
obs |
observed summary statistics. |
obs.scaled |
observed summary statistics standardized according to the mean and standard deviation of simulated values. |
ss |
standardized summary statistics of the communities simulated with parameter
values listed in |
ss.scale |
data frame including the mean and the standard deviation used for standardization of observed and summary statistics. |
abc |
a single (if |
F. Munoz, E. Barthelemy
Jabot, F., and J. Chave. 2009. Inferring the parameters of the neutral theory of biodiversity using phylogenetic information and implications for tropical forests. Ecology Letters 12:239-248.
Csillery, K., M. G. B. Blum, O. E. Gaggiotti, and O. Francois. 2010. Approximate Bayesian computation (ABC) in practice. Trends in Ecology & Evolution 25:410-418.
Csillery, K., O. Francois, and M. G. Blum. 2012. abc: an R package for Approximate Bayesian Computation (ABC). Methods in Ecology and Evolution 3:475-479.
Jabot, F., T. Faure, and N. Dumoulin 2013. EasyABC: performing efficient approximate Bayesian computation sampling schemes using R. Methods in Ecology and Evolution 4:684-687.
coalesc_abc_std(),
coalesc_easyABC(),
abc() in abc package,
parLapply() in parallel package.
## Not run: # 1/ Analysis of community assembly in a single community # Trait-dependent filtering function filt_gaussian <- function(t, par) exp(-(t - par[1])^2/(2*par[2]^2)) # Definition of parameters of this function and their range par.range <- data.frame(rbind(c(0, 1), c(0.05, 1))) row.names(par.range) <- c("topt", "sigmaopt") # Basic summary statistics describing local community composition f.sumstats <- function(com) array(dimnames=list(c("cwm", "cwv", "cws", "cwk", "S", "Es")), c(mean(com[,3]), var(com[,3]), e1071::skewness(com[,3]), e1071::kurtosis(com[,3]), vegan::specnumber(table(com[,2])), vegan::diversity(table(com[,2])))) # An observed community is simulated with known parameter values comm <- coalesc(J = 400, m = 0.5, theta = 50, filt = function(x) filt_gaussian(x, c(0.2, 0.1))) # ABC estimation of the parameters of filtering and migration based on observed # community composition # Number of values to sample in prior distributions nb.samp <- 1000 # Should be large ## Warning: this function may take a while system.time(res.single <- coalesc_abc(comm$com, comm$pool, f.sumstats = f.sumstats, filt.abc = filt_gaussian, par.filt = par.range, nb.samp = nb.samp, parallel = TRUE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet")) # If the filtering function can be applied directly to a vector of trait values, # it is strongly recommended to set "filt.vect = T" system.time(res.single <- coalesc_abc(comm$com, comm$pool, f.sumstats = f.sumstats, filt.abc = filt_gaussian, filt.vect = T, par.filt = par.range, nb.samp = nb.samp, parallel = TRUE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet")) plot(res.single$abc, param = res.single$par) hist(res.single$abc) # Cross validation ## Warning: this function is slow res.single$cv <- abc::cv4abc(param = res.single$par, sumstat = res.single$ss, nval = 100, tols = c(0.01, 0.1, 1), method.abc = "neuralnet") plot(res.single$cv) # Alternative option using a MCMC sampling approach res.single2 <- coalesc_abc(comm$com, comm$pool, f.sumstats=f.sumstats, filt.abc=filt_gaussian, par.filt=par.range, nb.samp=nb.samp/10, tol=0.1, type = "mcmc") # 2/ Analysis of community assembly in multiple communities with a common pool of # immigrants # When the input is a site-species matrix, use argument multi="tab" # When the input is a list of community composition, use argument multi="seqcom" # 2.1/ Environment-dependent environmental filtering # The variation of optimal trait values can depend on environmental variation across communities filt_gaussian_env <- function(t, par, env) exp(-(t-par[1]*env+par[2])^2/(2*par[3]^2)) # Vector of environmental conditions across communities env1 <- matrix(runif(20)) # Simulate a set of 20 communities with varying environmental filtering in # different environmental conditions # A common source pool of immigrants pool <- coalesc(J = 10000, theta = 50)$com meta1 <- c() for(i in 1:20) meta1[[i]] <- coalesc(J = 400, pool = pool, m = 0.5, filt = function(x) filt_gaussian_env(x, c(0.5, 0.2, 0.1), env1[i,])) # Build a species-by-site table tab1 <- array(0, c(20, max(pool$sp))) for(i in 1:20) { compo <- table(meta1[[i]]$com$sp) tab1[i, as.numeric(names(compo))] <- compo } colnames(tab1) <- as.character(1:ncol(tab1)) tab1 <- tab1[, colSums(tab1)!=0] # Definition of parameters and their range # We will estimate the slope a and intercept b of the linear relationship between # optimal trait values and environmental conditions par.range <- data.frame(rbind(c(-1, 1), c(0, 1), c(0.05, 1))) row.names(par.range) <- c("a", "b", "sigmaopt") # Basic summary statistics # The function will provide trait-based statistics and taxonomic diversity in # each community f.sumstats <- function(tab, traits) array(dimnames= list(c(paste(rep("cwm",nrow(tab)),1:nrow(tab)), paste(rep("cwv",nrow(tab)),1:nrow(tab)), paste(rep("cws",nrow(tab)),1:nrow(tab)), paste(rep("cwk",nrow(tab)),1:nrow(tab)), paste(rep("S",nrow(tab)),1:nrow(tab)), paste(rep("Es",nrow(tab)),1:nrow(tab)), paste(rep("beta",nrow(tab)),1:nrow(tab)))), c(apply(tab, 1, function(x) Weighted.Desc.Stat::w.mean(traits[colnames(tab),], x)), apply(tab, 1, function(x) Weighted.Desc.Stat::w.var(traits[colnames(tab),], x)), apply(tab, 1, function(x) Weighted.Desc.Stat::w.skewness(traits[colnames(tab),], x)), apply(tab, 1, function(x) Weighted.Desc.Stat::w.kurtosis(traits[colnames(tab),], x)), apply(tab, 1, function(x) sum(x!=0)), apply(tab, 1, vegan::diversity), colMeans(as.matrix(vegan::betadiver(tab,"w"))))) # 2.1.1/ Using multi=tab (no local trait information) # ABC estimation of the parameters of trait-environment relationship and of migration, # based on observed community composition ## Warning: this function may take a while res.tab1 <- coalesc_abc(tab1, pool, multi = "tab", f.sumstats = f.sumstats, filt.abc = filt_gaussian_env, filt.vect = T, par.filt = par.range, add = T, var.add = env1, nb.samp = nb.samp, parallel = TRUE, tol = 0.5, pkg = c("vegan","Weighted.Desc.Stat"), method.abc = "neuralnet") plot(res.tab1$abc, param=res.tab1$par) hist(res.tab1$abc) # 2.1.2/ Using multi=seqcom (keeping info on local trait values) seqcom <- lapply(meta1, function(x) x$com) f.sumstats.seqcom <- function(seqcom) unlist(lapply(seqcom, function(x) c(mean(x[,3]),var(x[,3]),length(unique(x[,2])), vegan::diversity(table(x[,2]))))) seqcom.obs <- f.sumstats.seqcom(seqcom) res.seqcom <- coalesc_abc(seqcom, pool, multi = "seqcom", f.sumstats = f.sumstats.seqcom, filt.abc = filt_gaussian_env, add = T, var.add = env1, prop = F, par.filt = par.range, nb.samp = nb.samp) # Another example with proportion data in communities seqcom.prop <- lapply(seqcom, function(x) data.frame( sp=tapply(x[,2], x[,2], function(y) y[1]), cov=tapply(x[,2], x[,2], length)/nrow(x), tr=tapply(x[,3], x[,2], mean))) # Default prior for community size is uniform between 100 and 1000 # Can be changed by setting par.size f.sumstats.seqcom.prop <- function(seqcom) unlist(lapply(seqcom, function(x) c(weighted.mean(x[,3], x[,2]),Weighted.Desc.Stat::w.var(x[,3], x[,2]),length(unique(x[,2])), sum(x[,2]^2)))) sumstats.obs <- f.sumstats.seqcom.prop(seqcom.prop) res.seqcom.prop <- coalesc_abc(seqcom.prop, pool, multi = "seqcom", f.sumstats = f.sumstats.seqcom.prop, filt.abc = filt_gaussian_env, add = T, var.add = env1, prop = T, par.filt = par.range, nb.samp = nb.samp) # 2.2/ Environment-dependent environmental filtering and immigration # In this case, the migration rate depends of another environmental variable # representing, e.g., community isolation # There will be two environmental variables used for parameter inference, one conditioning # environmental filtering, and the other conditioning migration env2 <- matrix(runif(20)) env <- cbind(env1, env2) colnames(env) <- c("env1", "env2") # Migration depends on environmental conditions following some linear relationship migr_env <- function(par, env) (par[2]-par[1])*env+par[1] # Simulate a set of 20 communities with environment-dependent environmental filtering # and immigration meta2 <- c() for(i in 1:20) meta2[[i]] <- coalesc(J = 400, pool = pool, m = migr_env(c(0.25,0.75), env[i,2]), filt = function(x) filt_gaussian_env(x, c(0.5, 0.2, 0.1), env[i,1])) # Build a species-by-site table tab2 <- array(0, c(20, max(pool$sp))) for(i in 1:20) { compo <- table(meta2[[i]]$com$sp) tab2[i, as.numeric(names(compo))] <- compo } colnames(tab2) <- as.character(1:ncol(tab2)) tab2 <- tab2[, colSums(tab2)!=0] # Definition of parameters and their range # We will estimate the slope a and intercept b of the linear relationship between # optimal trait values and environmental variable 1, and slope c and intercept d in the # logistic function determining the variation of migration rate with environmental variable 2 par.filt.range <- data.frame(rbind(c(-1, 1), c(0, 1), c(0.05, 1))) row.names(par.filt.range) <- c("a", "b", "sigmaopt") par.migr.range <- data.frame(rbind(c(0, 1), c(0, 1))) row.names(par.migr.range) <- c("c", "d") # ABC estimation of the parameters of trait-environment and migration-environment # relationships, based on observed community composition ## Warning: this function may take a while nb.samp <- 200 res.tab2 <- coalesc_abc(tab2, pool, multi = "tab", f.sumstats = f.sumstats, filt.abc = function(x, par, env) filt_gaussian_env(x, par, env[1]), filt.vect = T, migr.abc = function(par, env) migr_env(par, env[2]), par.filt = par.filt.range, par.migr = par.migr.range, add = T, var.add = env, nb.samp = nb.samp, parallel = FALSE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet") plot(res.tab2$abc, param=res.tab2$par) hist(res.tab2$abc) # Check result consistency comm <- 1 coeff <- summary(res.tab2$abc) plot(cbind(sapply(pool$tra, function(x) filt_gaussian_env(x, c(0.5, 0.2, 0.1), env[comm,1])), sapply(pool$tra, function(x) filt_gaussian_env(x, coeff[3,1:3], env[comm,1]))), xlab = paste("Expected filter value in community ", comm, sep=" "), ylab = "Estimated filter value") abline(0,1) plot(cbind(sapply(env[,2], function(x) migr_env(c(0.25, 0.75), x)), sapply(env[,2], function(x) migr_env(coeff[3,4:5], x))), xlab = "Expected migration value", ylab = "Estimated migration value") abline(0,1) # 3/ Distinct pools of migrants across communities # The immigrants can be drawn from distinct pools for each community, to represent a local context. # In this case, the pool argument sent to coalesc_abc must be a list of the local pools. # Simulate several communities with distinct pools and same environmental filter pool.loc <- c() meta3 <- c() theta <- 2.5*(1:20) # The pools differ in diversity # We first define a baseline pool of immigrants common to all communities baseline <- coalesc(J = 10000, theta = 25)$com baseline$sp <- paste(0, baseline$sp, sep = "_") for(i in 1:20) { pool.loc[[i]] <- coalesc(J = 10000, theta = theta[i])$com pool.loc[[i]]$sp <- paste(i, pool.loc[[i]]$sp, sep = "_") pool.loc[[i]] <- rbind(pool.loc[[i]], baseline) meta3[[i]] <- coalesc(J = 400, pool = pool.loc[[i]], m = 0.5, filt = function(x) filt_gaussian(x, c(0.25, 0.1))) } # Build a species-by-site table tab3 <- array(0, c(20, length(unique(Reduce(rbind, pool.loc)[,2])))) colnames(tab3) <- unique(Reduce(rbind, pool.loc)[,2]) for(i in 1:20) { compo <- table(meta3[[i]]$com$sp) tab3[i, names(compo)] <- compo } tab3 <- tab3[, colSums(tab3)!=0] # ABC estimation of the parameters of trait-environment and migration-environment # relationships, based on observed community composition ## Warning: this function may take a while par.range <- data.frame(rbind(c(0, 1), c(0.05, 1))) row.names(par.range) <- c("topt", "sigmaopt") res.tab3 <- coalesc_abc(tab3, pool.loc, multi = "tab", f.sumstats = f.sumstats, filt.abc = filt_gaussian, filt.vect = T, par.filt = par.range, nb.samp = nb.samp, parallel = FALSE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet") plot(res.tab3$abc, param=res.tab3$par) hist(res.tab3$abc) ## End(Not run)## Not run: # 1/ Analysis of community assembly in a single community # Trait-dependent filtering function filt_gaussian <- function(t, par) exp(-(t - par[1])^2/(2*par[2]^2)) # Definition of parameters of this function and their range par.range <- data.frame(rbind(c(0, 1), c(0.05, 1))) row.names(par.range) <- c("topt", "sigmaopt") # Basic summary statistics describing local community composition f.sumstats <- function(com) array(dimnames=list(c("cwm", "cwv", "cws", "cwk", "S", "Es")), c(mean(com[,3]), var(com[,3]), e1071::skewness(com[,3]), e1071::kurtosis(com[,3]), vegan::specnumber(table(com[,2])), vegan::diversity(table(com[,2])))) # An observed community is simulated with known parameter values comm <- coalesc(J = 400, m = 0.5, theta = 50, filt = function(x) filt_gaussian(x, c(0.2, 0.1))) # ABC estimation of the parameters of filtering and migration based on observed # community composition # Number of values to sample in prior distributions nb.samp <- 1000 # Should be large ## Warning: this function may take a while system.time(res.single <- coalesc_abc(comm$com, comm$pool, f.sumstats = f.sumstats, filt.abc = filt_gaussian, par.filt = par.range, nb.samp = nb.samp, parallel = TRUE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet")) # If the filtering function can be applied directly to a vector of trait values, # it is strongly recommended to set "filt.vect = T" system.time(res.single <- coalesc_abc(comm$com, comm$pool, f.sumstats = f.sumstats, filt.abc = filt_gaussian, filt.vect = T, par.filt = par.range, nb.samp = nb.samp, parallel = TRUE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet")) plot(res.single$abc, param = res.single$par) hist(res.single$abc) # Cross validation ## Warning: this function is slow res.single$cv <- abc::cv4abc(param = res.single$par, sumstat = res.single$ss, nval = 100, tols = c(0.01, 0.1, 1), method.abc = "neuralnet") plot(res.single$cv) # Alternative option using a MCMC sampling approach res.single2 <- coalesc_abc(comm$com, comm$pool, f.sumstats=f.sumstats, filt.abc=filt_gaussian, par.filt=par.range, nb.samp=nb.samp/10, tol=0.1, type = "mcmc") # 2/ Analysis of community assembly in multiple communities with a common pool of # immigrants # When the input is a site-species matrix, use argument multi="tab" # When the input is a list of community composition, use argument multi="seqcom" # 2.1/ Environment-dependent environmental filtering # The variation of optimal trait values can depend on environmental variation across communities filt_gaussian_env <- function(t, par, env) exp(-(t-par[1]*env+par[2])^2/(2*par[3]^2)) # Vector of environmental conditions across communities env1 <- matrix(runif(20)) # Simulate a set of 20 communities with varying environmental filtering in # different environmental conditions # A common source pool of immigrants pool <- coalesc(J = 10000, theta = 50)$com meta1 <- c() for(i in 1:20) meta1[[i]] <- coalesc(J = 400, pool = pool, m = 0.5, filt = function(x) filt_gaussian_env(x, c(0.5, 0.2, 0.1), env1[i,])) # Build a species-by-site table tab1 <- array(0, c(20, max(pool$sp))) for(i in 1:20) { compo <- table(meta1[[i]]$com$sp) tab1[i, as.numeric(names(compo))] <- compo } colnames(tab1) <- as.character(1:ncol(tab1)) tab1 <- tab1[, colSums(tab1)!=0] # Definition of parameters and their range # We will estimate the slope a and intercept b of the linear relationship between # optimal trait values and environmental conditions par.range <- data.frame(rbind(c(-1, 1), c(0, 1), c(0.05, 1))) row.names(par.range) <- c("a", "b", "sigmaopt") # Basic summary statistics # The function will provide trait-based statistics and taxonomic diversity in # each community f.sumstats <- function(tab, traits) array(dimnames= list(c(paste(rep("cwm",nrow(tab)),1:nrow(tab)), paste(rep("cwv",nrow(tab)),1:nrow(tab)), paste(rep("cws",nrow(tab)),1:nrow(tab)), paste(rep("cwk",nrow(tab)),1:nrow(tab)), paste(rep("S",nrow(tab)),1:nrow(tab)), paste(rep("Es",nrow(tab)),1:nrow(tab)), paste(rep("beta",nrow(tab)),1:nrow(tab)))), c(apply(tab, 1, function(x) Weighted.Desc.Stat::w.mean(traits[colnames(tab),], x)), apply(tab, 1, function(x) Weighted.Desc.Stat::w.var(traits[colnames(tab),], x)), apply(tab, 1, function(x) Weighted.Desc.Stat::w.skewness(traits[colnames(tab),], x)), apply(tab, 1, function(x) Weighted.Desc.Stat::w.kurtosis(traits[colnames(tab),], x)), apply(tab, 1, function(x) sum(x!=0)), apply(tab, 1, vegan::diversity), colMeans(as.matrix(vegan::betadiver(tab,"w"))))) # 2.1.1/ Using multi=tab (no local trait information) # ABC estimation of the parameters of trait-environment relationship and of migration, # based on observed community composition ## Warning: this function may take a while res.tab1 <- coalesc_abc(tab1, pool, multi = "tab", f.sumstats = f.sumstats, filt.abc = filt_gaussian_env, filt.vect = T, par.filt = par.range, add = T, var.add = env1, nb.samp = nb.samp, parallel = TRUE, tol = 0.5, pkg = c("vegan","Weighted.Desc.Stat"), method.abc = "neuralnet") plot(res.tab1$abc, param=res.tab1$par) hist(res.tab1$abc) # 2.1.2/ Using multi=seqcom (keeping info on local trait values) seqcom <- lapply(meta1, function(x) x$com) f.sumstats.seqcom <- function(seqcom) unlist(lapply(seqcom, function(x) c(mean(x[,3]),var(x[,3]),length(unique(x[,2])), vegan::diversity(table(x[,2]))))) seqcom.obs <- f.sumstats.seqcom(seqcom) res.seqcom <- coalesc_abc(seqcom, pool, multi = "seqcom", f.sumstats = f.sumstats.seqcom, filt.abc = filt_gaussian_env, add = T, var.add = env1, prop = F, par.filt = par.range, nb.samp = nb.samp) # Another example with proportion data in communities seqcom.prop <- lapply(seqcom, function(x) data.frame( sp=tapply(x[,2], x[,2], function(y) y[1]), cov=tapply(x[,2], x[,2], length)/nrow(x), tr=tapply(x[,3], x[,2], mean))) # Default prior for community size is uniform between 100 and 1000 # Can be changed by setting par.size f.sumstats.seqcom.prop <- function(seqcom) unlist(lapply(seqcom, function(x) c(weighted.mean(x[,3], x[,2]),Weighted.Desc.Stat::w.var(x[,3], x[,2]),length(unique(x[,2])), sum(x[,2]^2)))) sumstats.obs <- f.sumstats.seqcom.prop(seqcom.prop) res.seqcom.prop <- coalesc_abc(seqcom.prop, pool, multi = "seqcom", f.sumstats = f.sumstats.seqcom.prop, filt.abc = filt_gaussian_env, add = T, var.add = env1, prop = T, par.filt = par.range, nb.samp = nb.samp) # 2.2/ Environment-dependent environmental filtering and immigration # In this case, the migration rate depends of another environmental variable # representing, e.g., community isolation # There will be two environmental variables used for parameter inference, one conditioning # environmental filtering, and the other conditioning migration env2 <- matrix(runif(20)) env <- cbind(env1, env2) colnames(env) <- c("env1", "env2") # Migration depends on environmental conditions following some linear relationship migr_env <- function(par, env) (par[2]-par[1])*env+par[1] # Simulate a set of 20 communities with environment-dependent environmental filtering # and immigration meta2 <- c() for(i in 1:20) meta2[[i]] <- coalesc(J = 400, pool = pool, m = migr_env(c(0.25,0.75), env[i,2]), filt = function(x) filt_gaussian_env(x, c(0.5, 0.2, 0.1), env[i,1])) # Build a species-by-site table tab2 <- array(0, c(20, max(pool$sp))) for(i in 1:20) { compo <- table(meta2[[i]]$com$sp) tab2[i, as.numeric(names(compo))] <- compo } colnames(tab2) <- as.character(1:ncol(tab2)) tab2 <- tab2[, colSums(tab2)!=0] # Definition of parameters and their range # We will estimate the slope a and intercept b of the linear relationship between # optimal trait values and environmental variable 1, and slope c and intercept d in the # logistic function determining the variation of migration rate with environmental variable 2 par.filt.range <- data.frame(rbind(c(-1, 1), c(0, 1), c(0.05, 1))) row.names(par.filt.range) <- c("a", "b", "sigmaopt") par.migr.range <- data.frame(rbind(c(0, 1), c(0, 1))) row.names(par.migr.range) <- c("c", "d") # ABC estimation of the parameters of trait-environment and migration-environment # relationships, based on observed community composition ## Warning: this function may take a while nb.samp <- 200 res.tab2 <- coalesc_abc(tab2, pool, multi = "tab", f.sumstats = f.sumstats, filt.abc = function(x, par, env) filt_gaussian_env(x, par, env[1]), filt.vect = T, migr.abc = function(par, env) migr_env(par, env[2]), par.filt = par.filt.range, par.migr = par.migr.range, add = T, var.add = env, nb.samp = nb.samp, parallel = FALSE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet") plot(res.tab2$abc, param=res.tab2$par) hist(res.tab2$abc) # Check result consistency comm <- 1 coeff <- summary(res.tab2$abc) plot(cbind(sapply(pool$tra, function(x) filt_gaussian_env(x, c(0.5, 0.2, 0.1), env[comm,1])), sapply(pool$tra, function(x) filt_gaussian_env(x, coeff[3,1:3], env[comm,1]))), xlab = paste("Expected filter value in community ", comm, sep=" "), ylab = "Estimated filter value") abline(0,1) plot(cbind(sapply(env[,2], function(x) migr_env(c(0.25, 0.75), x)), sapply(env[,2], function(x) migr_env(coeff[3,4:5], x))), xlab = "Expected migration value", ylab = "Estimated migration value") abline(0,1) # 3/ Distinct pools of migrants across communities # The immigrants can be drawn from distinct pools for each community, to represent a local context. # In this case, the pool argument sent to coalesc_abc must be a list of the local pools. # Simulate several communities with distinct pools and same environmental filter pool.loc <- c() meta3 <- c() theta <- 2.5*(1:20) # The pools differ in diversity # We first define a baseline pool of immigrants common to all communities baseline <- coalesc(J = 10000, theta = 25)$com baseline$sp <- paste(0, baseline$sp, sep = "_") for(i in 1:20) { pool.loc[[i]] <- coalesc(J = 10000, theta = theta[i])$com pool.loc[[i]]$sp <- paste(i, pool.loc[[i]]$sp, sep = "_") pool.loc[[i]] <- rbind(pool.loc[[i]], baseline) meta3[[i]] <- coalesc(J = 400, pool = pool.loc[[i]], m = 0.5, filt = function(x) filt_gaussian(x, c(0.25, 0.1))) } # Build a species-by-site table tab3 <- array(0, c(20, length(unique(Reduce(rbind, pool.loc)[,2])))) colnames(tab3) <- unique(Reduce(rbind, pool.loc)[,2]) for(i in 1:20) { compo <- table(meta3[[i]]$com$sp) tab3[i, names(compo)] <- compo } tab3 <- tab3[, colSums(tab3)!=0] # ABC estimation of the parameters of trait-environment and migration-environment # relationships, based on observed community composition ## Warning: this function may take a while par.range <- data.frame(rbind(c(0, 1), c(0.05, 1))) row.names(par.range) <- c("topt", "sigmaopt") res.tab3 <- coalesc_abc(tab3, pool.loc, multi = "tab", f.sumstats = f.sumstats, filt.abc = filt_gaussian, filt.vect = T, par.filt = par.range, nb.samp = nb.samp, parallel = FALSE, tol = 0.5, pkg = c("e1071","vegan"), method.abc = "neuralnet") plot(res.tab3$abc, param=res.tab3$par) hist(res.tab3$abc) ## End(Not run)
Estimates parameters of neutral migration-drift dynamics (through migration
rate m and parameters of environmental filtering (through a filtering function
filt.abc()) from the composition of a local community and the related
regional pool.
coalesc_abc_std(comm.obs, pool = NULL, multi = "single", prop = F, traits = NULL, f.sumstats, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = T, size.abc = NULL, add = F, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = F, dim.pca = NULL, svd = F, theta.max = NULL, nb.samp = 10^6, parallel = TRUE, nb.core = NULL, tol = NULL, pkg = NULL, method.abc = "rejection") do.simul.coalesc(J, pool = NULL, multi = "single", prop = F, nb.com = NULL, traits = NULL, f.sumstats = NULL, nb.sumstats = NULL, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = T, size.abc = NULL, add = F, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, dim.pca = NULL, svd = F, theta.max = NULL, nb.samp = 10^6, parallel = TRUE, nb.core = NULL, pkg = NULL) generate_prior(pool = NULL, prop = F, constr = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, theta.max = NULL, nb.samp = 10^6)coalesc_abc_std(comm.obs, pool = NULL, multi = "single", prop = F, traits = NULL, f.sumstats, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = T, size.abc = NULL, add = F, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = F, dim.pca = NULL, svd = F, theta.max = NULL, nb.samp = 10^6, parallel = TRUE, nb.core = NULL, tol = NULL, pkg = NULL, method.abc = "rejection") do.simul.coalesc(J, pool = NULL, multi = "single", prop = F, nb.com = NULL, traits = NULL, f.sumstats = NULL, nb.sumstats = NULL, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = T, size.abc = NULL, add = F, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, dim.pca = NULL, svd = F, theta.max = NULL, nb.samp = 10^6, parallel = TRUE, nb.core = NULL, pkg = NULL) generate_prior(pool = NULL, prop = F, constr = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, theta.max = NULL, nb.samp = 10^6)
comm.obs |
the observed community composition. If |
pool |
composition of the regional pool to which the local community is
hypothesized to be related through migration dynamics with possible
environmental filtering. Should be a matrix of individuals on rows with
their individual id (first column), species id (second column), and
(optionally) the trait values of the individuals.
When |
multi |
structure of the community inputs:
|
prop |
indicates if the community composition is given in term of relative species abundances,
cover or proportions. In this case, a parameter of effective community size is estimated.
Default is FALSE. If no prior is provided for |
traits |
the trait values of species in the regional pool. It is used if trait
information is not provided in |
f.sumstats |
a function calculating the summary statistics of community
composition. It is be used to compare observed and simulated community
composition in ABC estimation. The first input argument of the function concerns community
composition, under a format depending on |
nb.sumstats |
number of summary statistics returned by |
filt.abc |
the hypothesized environmental filtering function. It is a function of
individual trait values (first argument), additional parameters to be estimated and (optionally)
environmental variables defined in |
filt.vect |
indicates whether the filtering function can be vectorized. It means that the function can take as input a vector of trait values and provide a vector of the corresponding weights. |
migr.abc |
a function defining immigration probability in local communities. It can be a function of
environmental variables defined in |
m.replace |
should the immigrants be drawn with replacement from the source pool. Default is |
size.abc |
a function defining local community sizes. It can be used when |
add |
indicates if additional variables must be passed to |
var.add |
additional variables to be passed to |
params |
equivalent to par.filt (see below): it is kept in the function for compatibility with previous versions. |
par.filt |
a matrix of the bounds of the parameters used in |
par.migr |
a matrix of the bounds of the parameters used in |
par.size |
a matrix of the bounds of the parameters used in |
constr |
constraints that must be set on parameter values, i.e., relationships that must
be met when drawing the values in prior distributions. It must be a vector of
character strings each including an operator relating some of the parameters.
The names of parameters must be consistent with those used in |
scale |
should the summary statistics be scaled. Default is |
dim.pca |
gives a number of independent dimensions to calculate from the simulated summary statistics,
by performing Principal Component Analysis (PCA). Default is |
svd |
indicates if Singular Value Decomposition (SVD) must be performed to get the basic
dimensions requested in |
theta.max |
if |
nb.samp |
the number of parameter values to be sampled in ABC calculation. Random
values of parameters of environmental filtering (see |
parallel |
boolean. If |
nb.core |
number of cores to be used in parallel computation if |
tol |
the tolerance value used in ABC estimation (see help in
|
pkg |
packages needed for calculation of |
J |
local community size(s). There must be several values when multi is |
nb.com |
number of communities. |
method.abc |
the method to be used in ABC estimation (see help on
|
coalesc_abc_std() provides ABC estimation of assembly parameters for one (if multi = "single",
default) or several communities (if multi = "tab" or "seqcom") related to the same
regional pool. It performs simulations with parameter values randomly drawn in priori distributions.
generate_prior() draws parameter values from the prior distributions defined with params,par.filt and/or par.size.
do.simul.coalesc() provides simulated communities used for ABC estimation.
par |
parameter values used in simulations. |
obs |
observed summary statistics. |
obs.scaled |
observed summary statistics standardized according to the mean and standard deviation of simulated values. |
ss |
standardized summary statistics of the communities simulated with parameter
values listed in |
ss.scale |
data frame including the mean and the standard deviation used for standardization of observed and summary statistics. |
abc |
a single (if |
F. Munoz, E. Barthelemy
Estimates parameters of neutral migration-drift dynamics (through migration
rate m and parameters of environmental filtering (through a filtering function
filt.abc()) from the composition of a local community and the related
regional pool.
coalesc_easyABC(comm.obs, pool = NULL, multi = "single", prop = F, traits = NULL, f.sumstats, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = T, size.abc = NULL, add = F, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = F, dim.pca = NULL, svd = F, theta.max = NULL, nb.samp = 10^6, parallel = TRUE, nb.core = NULL, tol = NULL, pkg = NULL, type = NULL, method.seq = "Lenormand", method.mcmc = "Marjoram_original", method.abc = "rejection", alpha = 0.5)coalesc_easyABC(comm.obs, pool = NULL, multi = "single", prop = F, traits = NULL, f.sumstats, filt.abc = NULL, filt.vect = F, migr.abc = NULL, m.replace = T, size.abc = NULL, add = F, var.add = NULL, params = NULL, par.filt = NULL, par.migr = NULL, par.size = NULL, constr = NULL, scale = F, dim.pca = NULL, svd = F, theta.max = NULL, nb.samp = 10^6, parallel = TRUE, nb.core = NULL, tol = NULL, pkg = NULL, type = NULL, method.seq = "Lenormand", method.mcmc = "Marjoram_original", method.abc = "rejection", alpha = 0.5)
comm.obs |
the observed community composition. If |
pool |
composition of the regional pool to which the local community is
hypothesized to be related through migration dynamics with possible
environmental filtering. Should be a matrix of individuals on rows with
their individual id (first column), species id (second column), and
(optionally) the trait values of the individuals.
When |
multi |
structure of the community inputs:
|
prop |
indicates if the community composition is given in term of relative species abundances,
cover or proportions. In this case, a parameter of effective community size is estimated.
Default is FALSE. If no prior is provided for |
traits |
the trait values of species in the regional pool. It is used if trait
information is not provided in |
f.sumstats |
a function calculating the summary statistics of community
composition. It is be used to compare observed and simulated community
composition in ABC estimation. The first input argument of the function concerns community
composition, under a format depending on |
filt.abc |
the hypothesized environmental filtering function. It is a function of
individual trait values (first argument), additional parameters to be estimated and (optionally)
environmental variables defined in |
filt.vect |
indicates whether the filtering function can be vectorized. It means that the function can take as input a vector of trait values and provide a vector of the corresponding weights. |
migr.abc |
a function defining immigration probability in local communities. It can be a function of
environmental variables defined in |
m.replace |
should the immigrants be drawn with replacement from the source pool. Default is |
size.abc |
a function defining local community sizes. It can be used when |
add |
indicates if additional variables must be passed to |
var.add |
additional variables to be passed to |
params |
equivalent to par.filt (see below): it is kept in the function for compatibility with previous versions. |
par.filt |
a matrix of the bounds of the parameters used in |
par.migr |
a matrix of the bounds of the parameters used in |
par.size |
a matrix of the bounds of the parameters used in |
constr |
constraints that must be set on parameter values, i.e., relationships that must
be met when drawing the values in prior distributions. It must be a vector of
character strings each including an operator relating some of the parameters.
The names of parameters must be consistent with those used in |
scale |
should the summary statistics be scaled. Default is |
dim.pca |
gives a number of independent dimensions to calculate from the simulated summary statistics,
by performing Principal Component Analysis (PCA). Default is |
svd |
indicates if Singular Value Decomposition (SVD) must be performed to get the basic
dimensions requested in |
theta.max |
if |
nb.samp |
the number of parameter values to be sampled in ABC calculation. Random
values of parameters of environmental filtering (see |
parallel |
boolean. If |
nb.core |
number of cores to be used in parallel computation if |
tol |
the tolerance value used in ABC estimation (see help in
|
pkg |
packages needed for calculation of |
type |
the type of algorithm to be used in EasyABC. Can be either "standard" (using package abc, default)" "seq" (sequential), "mcmc" or "annealing". Three later options are based on the EasyABC package. |
method.seq |
when type = "seq", gives the algorithm for sequential sampling scheme, which is passed to |
method.mcmc |
when type = "mcmc", gives the algorithm for MCMC sampling scheme, which is passed to |
method.abc |
the method to be used in ABC estimation (see help on
|
alpha |
a positive number between 0 and 1 (strictly) used when performing sequential ABC method. |
coalesc_easyABC() provides ABC estimation of assembly parameters for one (if multi = "single",
default) or several communities (if multi = "tab" or "seqcom") related to the same
regional pool. The function uses optimization algorithms implemented in the package EasyABC.
par |
parameter values used in simulations. |
obs |
observed summary statistics. |
obs.scaled |
observed summary statistics standardized according to the mean and standard deviation of simulated values. |
ss |
standardized summary statistics of the communities simulated with parameter
values listed in |
ss.scale |
data frame including the mean and the standard deviation used for standardization of observed and summary statistics. |
abc |
a single (if |
F. Munoz, E. Barthelemy
Jabot, F., T. Faure, and N. Dumoulin 2013. EasyABC: performing efficient approximate Bayesian computation sampling schemes using R. Methods in Ecology and Evolution 4:684-687.
Simulates niche-based (habitat filtering and/or limiting similarity) and neutral community dynamics from a given initial composition, over a given number of generations.
forward(initial, m = 1, theta = NULL, d = 1, gens = 150, keep = FALSE, pool = NULL, traits = NULL, filt = NULL, filt.vect = F, limit.sim = NULL, limit.intra = F, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", add = F, var.add = NULL, prob.death = NULL, method.dist = "euclidean", checks = T, plot_gens = FALSE) gauss_limit(dist, par) get_number_of_gens(given_size, pool, traits = NULL, nbrep = 5, m = 1, theta = NULL, d = 1, gens = NULL, filt = NULL, limit.sim = NULL, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", m.replace = T, add = F, var.add = NULL, prob.death = NULL, method.dist = "euclidean", plot_gens = FALSE) pick(com, d = 1, m = 1, theta = NULL, pool = NULL, prob.death = NULL, filt = NULL, filt.vect = F, limit.sim = NULL, limit.intra = F, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", m.replace = T, add = F, var.add = NULL, new.index = 0, method.dist = "euclidean") pick.mutate(com, d = 1, mu = 0, new.index = 0) pick.immigrate(com, d = 1, m = 1, pool, prob.death = NULL, filt = NULL, filt.vect = F, limit.sim = NULL, limit.intra = F, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", m.replace = T, add = F, var.add = NULL, method.dist = "euclidean")forward(initial, m = 1, theta = NULL, d = 1, gens = 150, keep = FALSE, pool = NULL, traits = NULL, filt = NULL, filt.vect = F, limit.sim = NULL, limit.intra = F, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", add = F, var.add = NULL, prob.death = NULL, method.dist = "euclidean", checks = T, plot_gens = FALSE) gauss_limit(dist, par) get_number_of_gens(given_size, pool, traits = NULL, nbrep = 5, m = 1, theta = NULL, d = 1, gens = NULL, filt = NULL, limit.sim = NULL, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", m.replace = T, add = F, var.add = NULL, prob.death = NULL, method.dist = "euclidean", plot_gens = FALSE) pick(com, d = 1, m = 1, theta = NULL, pool = NULL, prob.death = NULL, filt = NULL, filt.vect = F, limit.sim = NULL, limit.intra = F, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", m.replace = T, add = F, var.add = NULL, new.index = 0, method.dist = "euclidean") pick.mutate(com, d = 1, mu = 0, new.index = 0) pick.immigrate(com, d = 1, m = 1, pool, prob.death = NULL, filt = NULL, filt.vect = F, limit.sim = NULL, limit.intra = F, par.limit = 0.1, coeff.lim.sim = 1, type.filt = "immig", type.limit = "death", m.replace = T, add = F, var.add = NULL, method.dist = "euclidean")
com, initial
|
starting community. It is in principle a three (or more) column matrix or data.frame including individual ID, species names and trait values. For strictly neutral dynamics, it can be a vector of individual species names. |
m |
migration rate (if |
theta |
parameter of neutral dynamics in the regional pool (used only if |
mu |
mutation rate derived from |
d |
number of individuals that die in each time-step. |
gens |
number of generations to simulate. |
keep |
boolean value. If |
pool |
the regional pool of species providing immigrants to the local community. It is
in principle a three-column matrix or data frame including individual ID,
species names and trait values. If trait information is missing, a random trait
value is given to individuals, from a uniform distribution between 0 and 1.
If |
traits |
a matrix or data.frame including one or several traits on columns. A unique trait value is assigned to each species in the regional pool. Species names of |
given_size |
size of the community you want to have an estimate of the number of generations needed to reach stationarity in species richness. |
nbrep |
number of replicates from which you want to estimate the number of generations needed to reach stationarity in species richness. |
limit.sim |
if non null, limiting similarity will be simulated, based on species trait
distances (computed with the method given by |
limit.intra |
should limiting similarity play on individual dynamics within species? Default is |
par.limit |
a vector of additional parameters to be passed to the limiting similarity function defined in |
coeff.lim.sim |
adjust the intensity of limiting similarity. |
type.filt |
indicates how habitat filtering plays in community assembly.
If |
type.limit |
indicates how limiting similarity plays in community assembly.
If |
filt |
the function used to represent habitat filtering. For a given trait value
|
filt.vect |
indicates whether the filtering function can be vectorized. It means that the function can take as input a vector of trait values and provide a vector of the corresponding weights. |
m.replace |
should the immigrants be drawn with replacement from the source pool. Default is |
add |
indicates if additional variables must be passed to |
var.add |
additional variables to be passed to |
prob.death |
provides a baseline probability of death that is homogeneous across species. It is used in niche-based dynamics to represent the balance of baseline and niche-dependent mortality. |
method.dist |
provides the method to compute trait distances between individuals (syntax of
function |
new.index |
prefix used to give a new species name when speciation occurs. |
plot_gens |
plot the number of unique individuals and species over generations. |
checks |
should initial checks that the inputs are correct be performed? |
dist |
vector of trait distances that is used for calculation of limiting similarity. |
par |
addditional parameter values to be passed to |
The model of community assembly is a zero-sum game, so that the number of individuals of the community is fixed to the number of individuals in initial community.
Two types of niche-based processes, habitat filtering and limiting similarity
can affect (i) immigration, (ii) mortality, and (iii) recruitment of local offspring.
The user can define on each of the three components the processes play, by defining
type.filt and type.limit. These parameters can combine one of several
values in "immig", "death" and "loc.recr" depending on either the process plays on
immigration, (ii) mortality, and (iii) recruitment of local offspring, respectively.
For instance, defining type.filt = c("immig", "death") would mean that habitat
filtering plays on both immigration and on death events. By default, type.filt = "immig"
and type.simil = "death", which corresponds to a classical conception of
the hierarchical influence of habitat filtering and limiting similarity on immigration success
and local survival.
A environmental filtering function can be defined with filt. It should take two arguments, the first is a trait value of a candidate immigrant, the second is a vector including the parameter values of the filtering function. See examples below for further information on usage.
An important point is that in many cases the function might be vectorized, that is, it can provide a vector of filtering probabilities for a vector of trait values given in first argument. In this case the user should set filt.vect = T, which will significantly accelerate simulations.
gauss_limit() is the default function used to compute limiting similarity. In this case, the relative performance of an individual decreases depending on the sum of exponential distances to other individuals in the community (McArthur and Levins 1967), i.e.,
Function get_number_of_gen() allows determining the number of generations
needed to reach stationary richness for given parameterization of
forward(). The target number of generation is based on assessing the
change point in species richness change over time for replicate simulated
communities with random initial composition. A conservative measure is proposed
as the maximum time to reach stationary richness over the replicate simulated
communities.
Functions pick.immigrate() and pick.mutate() are used to simulate
immigration and speciation events within a time step. They are embedded in
forward and are not really intended for the end user.
A Shiny app is available to visualize simulated trait distributions for chosen parameter values in the model.
com |
if |
pool |
a data.frame of the individuals of the regional source pool, with the label of
ancestor individual in the regional pool on first column (as in first column of
input |
sp_t |
a vector of species richness at each time step. |
com_t |
if |
dist.t |
if |
new.index |
for |
call |
the call function. |
F. Munoz, derived from the untb function of R. Hankin.
For neutral dynamics, see S. P. Hubbell 2001. "The Unified Neutral Theory of Biodiversity". Princeton University Press.
For the default model of limiting similarity, MacArthur, R., Levins, R., 1967. The limiting similarity, convergence, and divergence of coexisting species. The American Naturalist 101, 377–385.
## Not run: # Initial community composed of 10 species each including 10 individuals initial1 <- rep(as.character(1:10), each = 10) # Simulation of speciation and drift dynamics over 1000 time steps final1 <- forward(initial = initial1, theta = 50, gens = 1000) # The final community includes new species (by default names begins with "new") final1$com$sp # includes new species generated by speciation events # A regional pool including 100 species each including 10 individuals pool <- rep(paste("pool.sp",as.character(1:100), sep=""), each = 10) # Simulation of migration and drift dynamics over 1000 time steps final2 <- forward(initial = initial1, m = 0.1, gens = 1000, pool = pool) # The final community includes species that have immigrated from the pool final2$com$sp # includes new species that immigrated from the pool # Initial community composed of 10 species each including 10 individuals, # with trait information for niche-based dynamics initial2 <- data.frame(ind = 1:100, sp = rep(as.character(1:10), each = 10), trait = runif(100), stringsAsFactors = F) # Simulation of stabilizing hab. filtering around t = 0.5, over 2000 time steps sigm <- 0.1 filt_gaussian <- function(t,x) exp(-(x - t)^2/(2*sigm^2)) # Filtering only plays only on immigration (default type.filt = "immig") final3 <- forward(initial = initial2, m = 0.1, gens = 2000, pool = pool, filt = function(x) filt_gaussian(0.5,x)) plot_comm(final3) # trait distribution in final community # Filtering also plays on local mortality and recruitment final4 <- forward(initial = initial2, m = 0.1, gens = 2000, pool = pool, filt = function(x) filt_gaussian(0.5,x), type.filt = c("immig", "death", "loc.recr")) plot_comm(final4) # Stronger convergence around 0.5 # Simulation of stabilizing hab. filtering with two traits pool <- data.frame(ind=1:10000, sp=rep(1:500, 20), tra1=rep(NA, 10000), tra2=rep(NA, 10000)) # Trait values in the pool t1.sp <- runif(500) t2.sp <- runif(500) pool[, 3] <- t1.sp[pool[,2]] pool[, 4] <- t2.sp[pool[,2]] initial3 <- pool[sample(1:10000, 100),] # Parameters of filtering for first trait topt1 <- 0.25 sigm1 <- 0.1 # Parameters of filtering for second trait topt2 <- 0.75 sigm2 <- 0.2 filt_gaussian <- function(x) exp(-(x[1] - topt1)^2/(2*sigm1^2)) * exp(-(x[2] - topt2)^2/(2*sigm2^2)) system.time(final5 <- forward(initial = initial3, m = 0.1, gens = 2000, pool = pool, filt = filt_gaussian)) # Here the filtering function can be vectorized system.time(final5 <- forward(initial = initial3, m = 0.1, gens = 2000, pool = pool, filt = filt_gaussian, filt.vect = T)) plot_comm(final5, seltrait = 1) # trait distribution in final community (first trait) plot_comm(final5, seltrait = 2) # trait distribution in final community (second trait) # Simulation of limiting similarity, over 2000 time steps final6 <- forward(initial = initial2, m = 0.1, gens = 2000, pool = pool, limit.sim = TRUE) plot_comm(final6) # Check temporal changes plot(final6$sp_t, xlab = "Time step", ylab = "Community richness") # Index of limiting similarity over time plot(final6$dist.t, xlab = "Time step", ylab = "Limiting similarity") # Higher migration rate, 5000 time steps # Limiting similarity only plays on mortality final7 <- forward(initial = initial2, m = 0.7, gens = 5000, pool = pool, limit.sim = TRUE) plot_comm(final7) # should be closer to regional distribution # Limiting similarity plays on immigration, mortality and local recruitment final8 <- forward(initial = initial2, m = 0.7, gens = 5000, pool = pool, limit.sim = TRUE, type.limit = c("immig", "death", "loc.recr")) plot_comm(final8) # Check temporal changes plot(final8$sp_t, xlab = "Time step", ylab = "Community richness") points(1:5000, final7$sp_t, col="blue") # Index of limiting similarity over time plot(final8$dist.t, xlab = "Time step", ylab = "Limiting similarity") points(1:5000, final7$dist.t, col="blue") ## End(Not run)## Not run: # Initial community composed of 10 species each including 10 individuals initial1 <- rep(as.character(1:10), each = 10) # Simulation of speciation and drift dynamics over 1000 time steps final1 <- forward(initial = initial1, theta = 50, gens = 1000) # The final community includes new species (by default names begins with "new") final1$com$sp # includes new species generated by speciation events # A regional pool including 100 species each including 10 individuals pool <- rep(paste("pool.sp",as.character(1:100), sep=""), each = 10) # Simulation of migration and drift dynamics over 1000 time steps final2 <- forward(initial = initial1, m = 0.1, gens = 1000, pool = pool) # The final community includes species that have immigrated from the pool final2$com$sp # includes new species that immigrated from the pool # Initial community composed of 10 species each including 10 individuals, # with trait information for niche-based dynamics initial2 <- data.frame(ind = 1:100, sp = rep(as.character(1:10), each = 10), trait = runif(100), stringsAsFactors = F) # Simulation of stabilizing hab. filtering around t = 0.5, over 2000 time steps sigm <- 0.1 filt_gaussian <- function(t,x) exp(-(x - t)^2/(2*sigm^2)) # Filtering only plays only on immigration (default type.filt = "immig") final3 <- forward(initial = initial2, m = 0.1, gens = 2000, pool = pool, filt = function(x) filt_gaussian(0.5,x)) plot_comm(final3) # trait distribution in final community # Filtering also plays on local mortality and recruitment final4 <- forward(initial = initial2, m = 0.1, gens = 2000, pool = pool, filt = function(x) filt_gaussian(0.5,x), type.filt = c("immig", "death", "loc.recr")) plot_comm(final4) # Stronger convergence around 0.5 # Simulation of stabilizing hab. filtering with two traits pool <- data.frame(ind=1:10000, sp=rep(1:500, 20), tra1=rep(NA, 10000), tra2=rep(NA, 10000)) # Trait values in the pool t1.sp <- runif(500) t2.sp <- runif(500) pool[, 3] <- t1.sp[pool[,2]] pool[, 4] <- t2.sp[pool[,2]] initial3 <- pool[sample(1:10000, 100),] # Parameters of filtering for first trait topt1 <- 0.25 sigm1 <- 0.1 # Parameters of filtering for second trait topt2 <- 0.75 sigm2 <- 0.2 filt_gaussian <- function(x) exp(-(x[1] - topt1)^2/(2*sigm1^2)) * exp(-(x[2] - topt2)^2/(2*sigm2^2)) system.time(final5 <- forward(initial = initial3, m = 0.1, gens = 2000, pool = pool, filt = filt_gaussian)) # Here the filtering function can be vectorized system.time(final5 <- forward(initial = initial3, m = 0.1, gens = 2000, pool = pool, filt = filt_gaussian, filt.vect = T)) plot_comm(final5, seltrait = 1) # trait distribution in final community (first trait) plot_comm(final5, seltrait = 2) # trait distribution in final community (second trait) # Simulation of limiting similarity, over 2000 time steps final6 <- forward(initial = initial2, m = 0.1, gens = 2000, pool = pool, limit.sim = TRUE) plot_comm(final6) # Check temporal changes plot(final6$sp_t, xlab = "Time step", ylab = "Community richness") # Index of limiting similarity over time plot(final6$dist.t, xlab = "Time step", ylab = "Limiting similarity") # Higher migration rate, 5000 time steps # Limiting similarity only plays on mortality final7 <- forward(initial = initial2, m = 0.7, gens = 5000, pool = pool, limit.sim = TRUE) plot_comm(final7) # should be closer to regional distribution # Limiting similarity plays on immigration, mortality and local recruitment final8 <- forward(initial = initial2, m = 0.7, gens = 5000, pool = pool, limit.sim = TRUE, type.limit = c("immig", "death", "loc.recr")) plot_comm(final8) # Check temporal changes plot(final8$sp_t, xlab = "Time step", ylab = "Community richness") points(1:5000, final7$sp_t, col="blue") # Index of limiting similarity over time plot(final8$dist.t, xlab = "Time step", ylab = "Limiting similarity") points(1:5000, final7$dist.t, col="blue") ## End(Not run)
Graphical function to used on the output of coalesc() or
forward() functions. It can show the links between regional and
local trait/abundance distributions, or local species and rank
abundance distributions.
plot_comm(x, type = "trait", seltrait = 1, main = NULL)plot_comm(x, type = "trait", seltrait = 1, main = NULL)
x |
a list including the species pool composition ( |
type |
|
seltrait |
index of the trait to be plotted following community data.frame (if multiple traits used in simulation). |
main |
an overall title for the plot. |
If type = "trait", the function provides density plots of the trait or
abundance distributions in the regional pool and in a local community.
If type = "locreg", it displays the relationship between
regional and local species relative abundances.
If type = "sad" or "rad", it shows the percentile plots provided by
functions in the "sads" package (see ppsad). The reference red line corresponds to log-series
for "sad" and to geometric series for "rad".
By default type = "trait".
To be used on the output of coalesc() or forward() functions.
Return two stacked ggplot2 density plots if
type = "trait" and biplots otherwise.
F. Munoz; P. Denelle
# Simulation of a neutral community including 100 individuals J <- 500; theta <- 50; m <- 0.1; comm1 <- coalesc(J, m, theta) plot_comm(comm1) plot_comm(comm1, type = "locreg") plot_comm(comm1, type = "sad") # Stabilizing habitat filtering around t = 0.5 filt_gaussian <- function(x) exp(-(x - 0.5)^2/(2*0.1^2)) comm2 <- coalesc(J, m, theta, filt = filt_gaussian) plot_comm(comm2) plot_comm(comm2, type = "locreg") plot_comm(comm1, type = "sad")# Simulation of a neutral community including 100 individuals J <- 500; theta <- 50; m <- 0.1; comm1 <- coalesc(J, m, theta) plot_comm(comm1) plot_comm(comm1, type = "locreg") plot_comm(comm1, type = "sad") # Stabilizing habitat filtering around t = 0.5 filt_gaussian <- function(x) exp(-(x - 0.5)^2/(2*0.1^2)) comm2 <- coalesc(J, m, theta, filt = filt_gaussian) plot_comm(comm2) plot_comm(comm2, type = "locreg") plot_comm(comm1, type = "sad")
This function performs posterior predicitve checks assessing a model's prediction performance of either user defined summary statistics, functional diversity (intrspecific trait distributions, moments of the trait distribution, Enquist et al., 2015) or species rank-abundance plots
prdch4coalesc(com.obs, pool, filt, params, stats = "abund", f.stats = NULL, estim = NULL, nval = 100, progress = TRUE)prdch4coalesc(com.obs, pool, filt, params, stats = "abund", f.stats = NULL, estim = NULL, nval = 100, progress = TRUE)
com.obs |
a data.frame of observed individuals, the first column should contain individual labels, the second the names of the species and the third species trait values - should be the same as provided for the ABC analysis in |
pool |
a data.frame containing the regional pool of species providing immigrants to the local community. It should include the label of individual on first column, its species on second column and their trait values in the third column as in |
filt |
the inferred environmental filtering function. If |
params |
the posterior distribution resulting from the ABC analysis, should be a data.frame containing each parameter distribution as columns (posterior distribution of the migration rate parameter ( |
stats |
statistics used for the predictive checks:
|
f.stats |
the user-defined function computing the summary statistics to be used in the predictive checks when |
estim |
an estimator of the posterior distribution - can be either |
nval |
the number of simulations to be run for computing predictive checks. If |
progress |
whether to display progress bar (default is |
pvalue |
probabilities that the observed user-defined statistics are greater than
the same statistics simulated by the model (if |
prd.moments |
probabilities that the first four moments of the observed trait distribution are greater than those simulated by the model (if |
underestim.sp |
the names of the under-represented species by the model evaluated by comparing observed relative abundances of species to those simulated by the model (if |
overestim.sp |
the names of over-represented species by the model evaluated by comparing observed relative abundances of species to those simulated by the model (if |
Also, if stats="abund", rank-abundance curves are displayed distinguishing between species whose relative abundance is either over or under-estimated by the model as well as the confidence interval of the rank-abundance curves derived from the simulated communities
E. Barthelemy
Enquist, Brian J., et al. "Scaling from traits to ecosystems: developing a general trait driver theory via integrating trait-based and metabolic scaling theories." Advances in ecological research. Vol. 52. Academic Press, 2015. 249-318.
Csillery, Katalin, et al. "Approximate Bayesian computation (ABC) in practice." Trends in ecology & evolution 25.7 (2010): 410-418.
pool <- data.frame(ind = 1:1000, sp = as.character(rep(1:50), each = 10), trait = runif(1000), stringsAsFactors = FALSE) com.obs <- pool[sample(nrow(pool), 50),] f.stats <- function(com){ tab <- table(com[,2]) as.vector(t(sapply(0:3, function(x) hillR::hill_taxa(tab, q=x)))) } filt <- function(x, par) exp(-(x - par[[1]])^2/(2*par[[2]]^2)) stats <- c("custom","abund", "moments") params <- data.frame(par1 = runif(100,0,1), par2 = runif(100, .5, .9), par3 = runif(100,0,1)) checks <- prdch4coalesc(com.obs, pool, filt, params, stats, f.stats)pool <- data.frame(ind = 1:1000, sp = as.character(rep(1:50), each = 10), trait = runif(1000), stringsAsFactors = FALSE) com.obs <- pool[sample(nrow(pool), 50),] f.stats <- function(com){ tab <- table(com[,2]) as.vector(t(sapply(0:3, function(x) hillR::hill_taxa(tab, q=x)))) } filt <- function(x, par) exp(-(x - par[[1]])^2/(2*par[[2]]^2)) stats <- c("custom","abund", "moments") params <- data.frame(par1 = runif(100,0,1), par2 = runif(100, .5, .9), par3 = runif(100,0,1)) checks <- prdch4coalesc(com.obs, pool, filt, params, stats, f.stats)
Adaptable statistic function to be provided into coalesc_abc2() to estimate parameters of neutral migration-drift dynamics and parameters of environmental filtering using ABC and adapted to the format imput of the local community. These return a set of diversity indexes which can be based on species local abundances, functional traits or both.
sumstats(com, multi="single", traits = NULL, type = "mix", n = 4, m = 4)sumstats(com, multi="single", traits = NULL, type = "mix", n = 4, m = 4)
com |
the observed community composition. If |
multi |
structure of the community input:
|
traits |
trait values (one or several traits in column) for the species present in |
type |
Type of diversity indices to be included in the summary statistics:
|
n |
number of Hill's diversity numbers to compute when |
m |
number of trait distribution moments to compute when |
sumstats provides summary statistics based on user preference and the format of community
data to be used in coalesc_abc2 for investigating neutral or niche-based dynamics in
community assembly using ABC.
A vector of summary statistics to be used to compare observed and simulated community composition in the ABC estimation performed by coalesc_abc2.
For community information of the type multi = "single" this vector is comprised of n Hill numbers (when type is "taxo", the m first moments of the trait distribution when type is "func" or the m first moments of the trait distribution followed by the n Hill numbers by when type is "mix" of the single community.
For community information as a species by site matrix or data.frame with X sites, this vector is comprised of n Hill numbers of the first community, followed by the n Hill numbers for the second and so on until X (when type is "taxo"). Likewise (when type is "func", with m first moments of the trait distributions of the X communities. When type is "mix", the vector gives the m first moments of the trait distribution for every community followed by their n Hill numbers.
E.Barthelemy
Ref?
## Not run: examples? ## End(Not run)## Not run: examples? ## End(Not run)
Create two random vectors of traits correlated between each other or a vector
of traits correlated to an existing one. The linear correlation is
defined by the parameter rho.
tcor(n, rho = 0.5, mar.fun = rnorm, x = NULL, ...)tcor(n, rho = 0.5, mar.fun = rnorm, x = NULL, ...)
n |
the integer number of values to be generated. |
rho |
a numeric parameter defining the linear correlation between the two traits (default is 0.5). It must belong to the interval [-1, 1]. |
x |
an vector of numeric values. Default is NULL. |
mar.fun |
a function defining the random generation for the trait distribution. Default is
|
... |
other arguments for the |
rho parameter is set to 0.5 by default. x = NULL by default.
Code adapted from: http://stats.stackexchange.com/questions/15011/generate-a-random-variable-with-a-defined-correlation-to-an-existing-variable
Return a data.frame with two numeric columns, each column defining a trait.
P. Denelle F. Munoz
# With no predefined trait traits <- tcor(n = 10000, rho = 0.8) plot(traits[, 1], traits[, 2]) cor(traits[, 1], traits[, 2]) # With existing trait existing_trait <- rnorm(10000, 10, 1) traits <- tcor(n = 10000, rho = 0.8, x = existing_trait) plot(traits[, 1], traits[, 2]) cor(traits[, 1], traits[, 2])# With no predefined trait traits <- tcor(n = 10000, rho = 0.8) plot(traits[, 1], traits[, 2]) cor(traits[, 1], traits[, 2]) # With existing trait existing_trait <- rnorm(10000, 10, 1) traits <- tcor(n = 10000, rho = 0.8, x = existing_trait) plot(traits[, 1], traits[, 2]) cor(traits[, 1], traits[, 2])