context("Unit-testing the basic likelihood calculations in BioGeoBEARS and the testthat package")
#
# Following example testthat usages at e.g.
# http://kbroman.org/pkg_primer/pages/tests.html
#
test_that(desc="Check that cladoRcpp version number is >= 0.15", code={
version_number = packageVersion("cladoRcpp")
TF = version_number >= 0.15
if (TF == FALSE)
{
txt = 'STOP ERROR inside test_that(desc="Check that cladoRcpp version number is >= 0.15"): the BioGeoBEARS "testthat" tests, located in BioGeoBEARS/tests, require that cladoRcpp have version 0.15 or higher to work. To get the new version, try "devtools::install_github(repo="nmatzke/cladoRcpp", quick=TRUE, dependencies=FALSE, build_vignettes=FALSE, keep_source=TRUE, local=FALSE, force=TRUE)".'
cat("\n\n")
cat(txt)
cat("\n\n")
}
expect_equal(object=TF, expected=TRUE)
}) # END test_that
test_that(desc="Check that phytools is installed", code={
TF = is.element("phytools", installed.packages()[,1])
if (TF == FALSE)
{
txt = 'STOP ERROR inside test_that(desc="Check that phytools is installed"): the BioGeoBEARS "testthat" tests, located in BioGeoBEARS/tests, require that phytools be installed. To get it, try \n\ninstall.packages("devtools")\n.\n'
cat("\n\n")
cat(txt)
cat("\n\n")
}
expect_equal(object=TF, expected=TRUE)
}) # END test_that
test_that(desc="Check BioGeoBEARS Mk model (a modified BAYAREALIKE+a model) against the phytools R package's Mk model, on Hawaiian Psychotria dataset (Ree & Smith 2008).", code={
# Skip the slow tests in online checks
testthat::skip_on_cran()
testthat::skip_on_travis()
#######################################################
# phytools ancestral states (ER model)
#######################################################
library(ape)
library(phytools)
library(BioGeoBEARS)
# Get phylogeny:
extdata_dir = np(system.file("extdata", package="BioGeoBEARS"))
trfn = np(paste(addslash(extdata_dir), "Psychotria_5.2.newick", sep=""))
tr = read.tree(trfn)
# This is the example geography file for Hawaiian Psychotria
# (from Ree & Smith 2008)
geogfn = np(paste(addslash(extdata_dir), "Psychotria_geog.data", sep=""))
tipnodenums = 1:length(tr$tip.label)
rootnodenum = length(tr$tip.label) + 1
intnodenums = (length(tr$tip.label)+1):(length(tr$tip.label) + tr$Nnode)
intnodenums
# Look at your geographic range data:
tipranges = getranges_from_LagrangePHYLIP(lgdata_fn=geogfn)
tipranges
areanames = names(tipranges@df)
areanums = 1:length(areanames)
tipvals = rep(NA, nrow(tipranges@df))
for (i in 1:nrow(tipranges@df))
{
tipvals[i] = areanums[unlist(tipranges@df[i,])==1]
}
names(tipvals) = row.names(tipranges@df)
tipvals
phytools_res = rerootingMethod(tree=tr, x=tipvals, model="ER")
phytools_res
phytools_ML = phytools_res$marginal.anc
library(cladoRcpp)
#source("/drives/Dropbox/_njm/__packages/cladoRcpp_setup/cladoRcpp.R")
library(BioGeoBEARS)
library(parallel) # for detectCores
library(GenSA)
# Set your working directory for output files
# default here is your home directory ("~")
# Change this as you like
wd = np("~")
setwd(wd)
# Double-check your working directory with getwd()
getwd()
# However, you can find the extdata directory like this:
extdata_dir = np(system.file("extdata", package="BioGeoBEARS"))
extdata_dir
#######################################################
# This is the example Newick file for Hawaiian Psychotria
# (from Ree & Smith 2008)
# "trfn" = "tree file name"
trfn = np(paste(addslash(extdata_dir), "Psychotria_5.2.newick", sep=""))
# Look at your phylogeny:
tr = read.tree(trfn)
# This is the example geography file for Hawaiian Psychotria
# (from Ree & Smith 2008)
geogfn = np(paste(addslash(extdata_dir), "Psychotria_geog.data", sep=""))
# Look at your geographic range data:
tipranges = getranges_from_LagrangePHYLIP(lgdata_fn=geogfn)
tipranges
# Maximum range size observed:
max(rowSums(dfnums_to_numeric(tipranges@df)))
# Set the maximum number of areas any species may occupy; this cannot be larger
# than the number of areas you set up, but it can be smaller.
max_range_size = 1
#######################################################
# Run BAYAREALIKE
#######################################################
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
BioGeoBEARS_run_object$trfn = trfn
BioGeoBEARS_run_object$geogfn = geogfn
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = FALSE
# Set up a time-stratified analysis:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = "GenSA" # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
#######################################################
# The Mk-model in BioGeoBEARS, i.e.,
# The BAYAREALIKE model, but allowing ranges of only size 1
# (no null range either)
# -d, -e (range expansion and range contraction play no role
# in Mk, a single-areas-allowed model
# +a (rate of Anagenetic range-switching)
# No cladogeneis process except y (range-copying/single-area
# narrow sympatry)
#######################################################
# Set up BAYAREALIKE model
# No subset sympatry
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","est"] = 0.0
# No vicariance
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","est"] = 0.0
# No jump dispersal/founder-event speciation
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","type"] = "free"
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","init"] = 0.01
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","est"] = 0.01
# Adjust linkage between parameters
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ysv","type"] = "1-j"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ys","type"] = "ysv*1/1"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["y","type"] = "1-j"
# Only sympatric/range-copying (y) events allowed, and with
# exact copying (both descendants always the same size as the ancestor)
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","init"] = 0.9999
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","est"] = 0.9999
# Turn off d and e; turn on a
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","est"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","est"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["a","type"] = "free"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["a","init"] = 0.03
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["a","est"] = 0.03
# Check the inputs
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
runslow = TRUE
resfn = "Psychotria_BAYAREALIKE_M0_unconstrained_v1.Rdata"
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resBAYAREALIKEa = res
} else {
# Loads to "res"
load(resfn)
resBAYAREALIKEa = res
}
#######################################################
# Store phytools results in a BioGeoBEARS-type format
#######################################################
resBAYAREALIKEcopy = resBAYAREALIKEa
resBAYAREALIKEcopy$ML_marginal_prob_each_state_at_branch_top_AT_node[intnodenums, ] = phytools_ML
#######################################################
# START VALIDATION CHECK
#######################################################
# Log-likelihood DOES match:
phytools_res$loglik
# -24.36196
resBAYAREALIKEa$total_loglikelihood
# -22.97566
# Lagrange/BioGeoBEARS leave out the root state frequencies
# phytools includes root state frequencies
lnL_root_states = sum(log(0.25))
lnL_root_states
resBAYAREALIKEa$total_loglikelihood + lnL_root_states
# -24.36196
# Matches!
# The rate parameter DOES match
phytools_res$Q[1,2]
# 0.07572288
resBAYAREALIKEa$outputs@params_table["a","est"]
# 0.07575708
#######################################################
# END VALIDATION CHECK
#######################################################
# Check lnLs
expect_equal(object=round(resBAYAREALIKEa$total_loglikelihood + lnL_root_states, digits=4), expected=round(phytools_res$loglik, digits=4))
cat("\n")
txt = "Checking the BioGeoBEARS lnL (+log 0.25 for base frequencies) under Mk model == phytools lnL under Mk model on the Psychotria dataset..."
cat(txt)
cat("...PASSED ")
# Check root state probabilities
rootnode = length(tr$tip.label) + 1
BGB_root_probs = round(x=resBAYAREALIKEa$ML_marginal_prob_each_state_at_branch_top_AT_node[rootnode,], digits=2)
phytools_root_probs = round(x=resBAYAREALIKEcopy$ML_marginal_prob_each_state_at_branch_top_AT_node[rootnode,], digits=2)
BGB_root_probs
phytools_root_probs
cat("\n")
txt = "Checking the BioGeoBEARS root state probabilities under Mk model == phytools root state probabilities under Mk model on the Psychotria dataset..."
cat(txt)
expect_equal(object=BGB_root_probs, expected=phytools_root_probs)
cat("...PASSED ")
# Check all ancestral state probabilities
phytools_ancprobs_allstates = round(resBAYAREALIKEcopy$ML_marginal_prob_each_state_at_branch_top_AT_node, digits=2)
BGB_ancprobs_allstates = round(resBAYAREALIKEa$ML_marginal_prob_each_state_at_branch_top_AT_node, digits=2)
cat("\n")
txt = "Checking that ALL BioGeoBEARS ancestral state probabilities under Mk model == all phytools ancestral state probabilities under Mk model on the Psychotria dataset..."
cat(txt)
expect_equal(object=BGB_ancprobs_allstates, expected=phytools_ancprobs_allstates)
cat("...PASSED ")
# Check that the inferred rate parameter matches
phytools_rate = round(phytools_res$Q[1,2], digits=3)
BGB_rate = round(resBAYAREALIKEa$outputs@params_table["a","est"], digits=3)
cat("\n")
txt = "Checking that BioGeoBEARS inferred rate under Mk model == phytools inferred rate under Mk model on the Psychotria dataset..."
cat(txt)
expect_equal(object=BGB_rate, expected=phytools_rate)
cat("...PASSED ")
cat("\n\n")
txt = "CHECKS PASSED: LIKELIHOODS AND ANCESTRAL STATE PROBABILITIES MATCH BETWEEN BioGeoBEARS BAYAREALIKE+a MODEL AND phytools Mk model. Test script from BioGeoBEARS validation page: http://phylo.wikidot.com/biogeobears-validation#phytools_vs_ancprobs_unconstr"
cat(txt)
cat("\n\n")
}) # END test_that
test_that(desc="Check BioGeoBEARS DEC model ML and parameters against Lagrange (M0) with Hawaiian Psychotria dataset (Ree & Smith 2008).", code={
# Skip the slow tests in online checks
testthat::skip_on_cran()
testthat::skip_on_travis()
library(cladoRcpp)
#source("/drives/Dropbox/_njm/__packages/cladoRcpp_setup/cladoRcpp.R")
library(BioGeoBEARS)
library(parallel) # for detectCores
# Set your working directory for output files
# default here is your home directory ("~")
# Change this as you like
wd = np("~")
setwd(wd)
# Double-check your working directory with getwd()
getwd()
#######################################################
# SETUP: Extension data directory
#######################################################
# When R packages contain extra files, they are stored in the "extdata" directory
# inside the installed package.
#
# BioGeoBEARS contains various example files and scripts in its extdata directory.
#
# Each computer operating system might install BioGeoBEARS in a different place,
# depending on your OS and settings.
#
# However, you can find the extdata directory like this:
extdata_dir = np(system.file("extdata", package="BioGeoBEARS"))
extdata_dir
list.files(extdata_dir)
# "system.file" looks in the directory of a specified package (in this case BioGeoBEARS)
# The function "np" is just a shortcut for normalizePath(), which converts the
# path to the format appropriate for your system (e.g., Mac/Linux use "/", but
# Windows uses "\\", if memory serves).
# Even when using your own data files, you should KEEP these commands in your
# script, since the plot_BioGeoBEARS_results function needs a script from the
# extdata directory to calculate the positions of "corners" on the plot. This cannot
# be made into a straight up BioGeoBEARS function because it uses C routines
# from the package APE which do not pass R CMD check for some reason.
#######################################################
# SETUP: YOUR TREE FILE AND GEOGRAPHY FILE
#######################################################
# Example files are given below. To run your own data,
# make the below lines point to your own files, e.g.
# trfn = "/mydata/frogs/frogBGB/tree.newick"
# geogfn = "/mydata/frogs/frogBGB/geog.data"
#######################################################
# Phylogeny file
# Notes:
# 1. Must be binary/bifurcating: no polytomies
# 2. No negative branchlengths (e.g. BEAST MCC consensus trees sometimes have negative branchlengths)
# 3. Be careful of very short branches, as BioGeoBEARS will interpret ultrashort branches as direct ancestors
# 4. You can use non-ultrametric trees, but BioGeoBEARS will interpret any tips significantly below the
# top of the tree as fossils! This is only a good idea if you actually do have fossils in your tree,
# as in e.g. Wood, Matzke et al. (2013), Systematic Biology.
# 5. The default settings of BioGeoBEARS make sense for trees where the branchlengths are in units of
# millions of years, and the tree is 1-1000 units tall. If you have a tree with a total height of
# e.g. 0.00001, you will need to adjust e.g. the max values of d and e, or (simpler) multiply all
# your branchlengths to get them into reasonable units.
# 6. DON'T USE SPACES IN SPECIES NAMES, USE E.G. "_"
#######################################################
# This is the example Newick file for Hawaiian Psychotria
# (from Ree & Smith 2008)
# "trfn" = "tree file name"
trfn = np(paste(addslash(extdata_dir), "Psychotria_5.2.newick", sep=""))
# Look at your phylogeny:
tr = read.tree(trfn)
#######################################################
# Geography file
# Notes:
# 1. This is a PHYLIP-formatted file. This means that in the
# first line,
# - the 1st number equals the number of rows (species)
# - the 2nd number equals the number of columns (number of areas)
# - after a tab, put the areas in parentheses, with spaces: (A B C D)
#
# 1.5. Example first line:
# 10 4 (A B C D)
#
# 2. The second line, and subsequent lines:
# speciesA 0110
# speciesB 0111
# speciesC 0001
# ...
#
# 2.5a. This means a TAB between the species name and the area 0/1s
# 2.5b. This also means NO SPACE AND NO TAB between the area 0/1s.
#
# 3. See example files at:
# http://phylo.wikidot.com/biogeobears#files
#
# 4. Make you understand what a PLAIN-TEXT EDITOR is:
# http://phylo.wikidot.com/biogeobears#texteditors
#
# 3. The PHYLIP format is the same format used for C++ LAGRANGE geography files.
#
# 4. All names in the geography file must match names in the phylogeny file.
#
# 5. DON'T USE SPACES IN SPECIES NAMES, USE E.G. "_"
#
# 6. Operational taxonomic units (OTUs) should ideally be phylogenetic lineages,
# i.e. genetically isolated populations. These may or may not be identical
# with species. You would NOT want to just use specimens, as each specimen
# automatically can only live in 1 area, which will typically favor DEC+J
# models. This is fine if the species/lineages really do live in single areas,
# but you wouldn't want to assume this without thinking about it at least.
# In summary, you should collapse multiple specimens into species/lineages if
# data indicates they are the same genetic population.
######################################################
# This is the example geography file for Hawaiian Psychotria
# (from Ree & Smith 2008)
geogfn = np(paste(addslash(extdata_dir), "Psychotria_geog.data", sep=""))
# Look at your geographic range data:
tipranges = getranges_from_LagrangePHYLIP(lgdata_fn=geogfn)
tipranges
# Maximum range size observed:
max(rowSums(dfnums_to_numeric(tipranges@df)))
# Set the maximum number of areas any species may occupy; this cannot be larger
# than the number of areas you set up, but it can be smaller.
max_range_size = 4
#######################################################
#######################################################
# DEC AND DEC+J ANALYSIS
#######################################################
#######################################################
# NOTE: The BioGeoBEARS "DEC" model is identical with
# the Lagrange DEC model, and should return identical
# ML estimates of parameters, and the same
# log-likelihoods, for the same datasets.
#
# Ancestral state probabilities at nodes will be slightly
# different, since BioGeoBEARS is reporting the
# ancestral state probabilities under the global ML
# model, and Lagrange is reporting ancestral state
# probabilities after re-optimizing the likelihood
# after fixing the state at each node. These will
# be similar, but not identical. See Matzke (2014),
# Systematic Biology, for discussion.
#
# Also see Matzke (2014) for presentation of the
# DEC+J model.
#######################################################
#######################################################
#######################################################
#######################################################
#######################################################
# Run DEC
#######################################################
# Intitialize a default model (DEC model)
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
# Give BioGeoBEARS the location of the phylogeny Newick file
BioGeoBEARS_run_object$trfn = trfn
# Give BioGeoBEARS the location of the geography text file
BioGeoBEARS_run_object$geogfn = geogfn
# Input the maximum range size
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
# 1. Here, un-comment ONLY the files you want to use.
# 2. Also un-comment "BioGeoBEARS_run_object = section_the_tree(...", below.
# 3. For example files see (a) extdata_dir,
# or (b) http://phylo.wikidot.com/biogeobears#files
# and BioGeoBEARS Google Group posts for further hints)
#
# Uncomment files you wish to use in time-stratified analyses:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = FALSE # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
# (use more cores to speed it up; this requires
# library(parallel) and/or library(snow). The package "parallel"
# is now default on Macs in R 3.0+, but apparently still
# has to be typed on some Windows machines. Note: apparently
# parallel works on Mac command-line R, but not R.app.
# BioGeoBEARS checks for this and resets to 1
# core with R.app)
# Sparse matrix exponentiation is an option for huge numbers of ranges/states (600+)
# I have experimented with sparse matrix exponentiation in EXPOKIT/rexpokit,
# but the results are imprecise and so I haven't explored it further.
# In a Bayesian analysis, it might work OK, but the ML point estimates are
# not identical.
# Also, I have not implemented all functions to work with force_sparse=TRUE.
# Volunteers are welcome to work on it!!
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up DEC model
# (nothing to do; defaults)
# Look at the BioGeoBEARS_run_object; it's just a list of settings etc.
BioGeoBEARS_run_object
# This contains the model object
BioGeoBEARS_run_object$BioGeoBEARS_model_object
# This table contains the parameters of the model
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table
# Run this to check inputs. Read the error messages if you get them!
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
# For a slow analysis, run once, then set runslow=FALSE to just
# load the saved result.
runslow = TRUE
resfn = "Psychotria_DEC_M0_unconstrained_v1.Rdata"
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resDEC = res
} else {
# Loads to "res"
load(resfn)
resDEC = res
}
resDEC
# Python (2013) result
# -lnL = 34.54
# dispersal = 0.03505
# extinction = 0.02831
# C++ Lagrange result
# dis: 0.0350117 ext: 0.0282904
# final -ln likelihood: 34.542
d = resDEC$output@params_table["d", "est"]
e = resDEC$output@params_table["e", "est"]
lnL = resDEC$total_loglikelihood
cat("\n")
txt = "Checking that BioGeoBEARS inferred DEC lnL == Lagrange DEC (Python or C++) lnL on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=lnL, digits=2), expected=round(x=-34.54196, digits=2))
cat("...PASSED ")
cat("\n")
txt = "Checking that BioGeoBEARS ML inferred 'd' DEC == Lagrange DEC (Python or C++) 'd' on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=d, digits=3), expected=round(x=0.03504546, digits=3))
cat("...PASSED ")
cat("\n")
txt = "Checking that BioGeoBEARS ML inferred 'e' DEC == Lagrange DEC (Python or C++) 'e' on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=e, digits=3), expected=round(x=0.02831871, digits=3))
cat("...PASSED ")
cat("\n\n")
txt = "CHECKS PASSED: Likelihoods and ML parameter estimates match the Lagrange (Python or C++) DEC results on the Psychotria dataset; see the original 'out-of-the-box' example script at http://phylo.wikidot.com/biogeobears#script"
cat(txt)
cat("\n\n")
}) # END test_that
test_that(desc="Check BioGeoBEARS DEC model ML and parameters against Lagrange (M0, but time-stratified) with Hawaiian Psychotria dataset (Ree & Smith 2008).", code={
# Skip the slow tests in online checks
testthat::skip_on_cran()
testthat::skip_on_travis()
library(cladoRcpp)
#source("/drives/Dropbox/_njm/__packages/cladoRcpp_setup/cladoRcpp.R")
library(BioGeoBEARS)
library(parallel) # for detectCores
library(GenSA)
# Set your working directory for output files
# default here is your home directory ("~")
# Change this as you like
wd = np("~")
setwd(wd)
# Double-check your working directory with getwd()
getwd()
#######################################################
# SETUP: Extension data directory
#######################################################
# When R packages contain extra files, they are stored in the "extdata" directory
# inside the installed package.
#
# BioGeoBEARS contains various example files and scripts in its extdata directory.
#
# Each computer operating system might install BioGeoBEARS in a different place,
# depending on your OS and settings.
#
# However, you can find the extdata directory like this:
extdata_dir = np(system.file("extdata", package="BioGeoBEARS"))
extdata_dir
list.files(extdata_dir)
# "system.file" looks in the directory of a specified package (in this case BioGeoBEARS)
# The function "np" is just a shortcut for normalizePath(), which converts the
# path to the format appropriate for your system (e.g., Mac/Linux use "/", but
# Windows uses "\\", if memory serves).
# Even when using your own data files, you should KEEP these commands in your
# script, since the plot_BioGeoBEARS_results function needs a script from the
# extdata directory to calculate the positions of "corners" on the plot. This cannot
# be made into a straight up BioGeoBEARS function because it uses C routines
# from the package APE which do not pass R CMD check for some reason.
#######################################################
# SETUP: YOUR TREE FILE AND GEOGRAPHY FILE
#######################################################
# Example files are given below. To run your own data,
# make the below lines point to your own files, e.g.
# trfn = "/mydata/frogs/frogBGB/tree.newick"
# geogfn = "/mydata/frogs/frogBGB/geog.data"
#######################################################
# Phylogeny file
# Notes:
# 1. Must be binary/bifurcating: no polytomies
# 2. No negative branchlengths (e.g. BEAST MCC consensus trees sometimes have negative branchlengths)
# 3. Be careful of very short branches, as BioGeoBEARS will interpret ultrashort branches as direct ancestors
# 4. You can use non-ultrametric trees, but BioGeoBEARS will interpret any tips significantly below the
# top of the tree as fossils! This is only a good idea if you actually do have fossils in your tree,
# as in e.g. Wood, Matzke et al. (2013), Systematic Biology.
# 5. The default settings of BioGeoBEARS make sense for trees where the branchlengths are in units of
# millions of years, and the tree is 1-1000 units tall. If you have a tree with a total height of
# e.g. 0.00001, you will need to adjust e.g. the max values of d and e, or (simpler) multiply all
# your branchlengths to get them into reasonable units.
# 6. DON'T USE SPACES IN SPECIES NAMES, USE E.G. "_"
#######################################################
# This is the example Newick file for Hawaiian Psychotria
# (from Ree & Smith 2008)
# "trfn" = "tree file name"
trfn = np(paste(addslash(extdata_dir), "Psychotria_5.2.newick", sep=""))
# Look at your phylogeny:
tr = read.tree(trfn)
#######################################################
# Geography file
# Notes:
# 1. This is a PHYLIP-formatted file. This means that in the
# first line,
# - the 1st number equals the number of rows (species)
# - the 2nd number equals the number of columns (number of areas)
# - after a tab, put the areas in parentheses, with spaces: (A B C D)
#
# 1.5. Example first line:
# 10 4 (A B C D)
#
# 2. The second line, and subsequent lines:
# speciesA 0110
# speciesB 0111
# speciesC 0001
# ...
#
# 2.5a. This means a TAB between the species name and the area 0/1s
# 2.5b. This also means NO SPACE AND NO TAB between the area 0/1s.
#
# 3. See example files at:
# http://phylo.wikidot.com/biogeobears#files
#
# 4. Make you understand what a PLAIN-TEXT EDITOR is:
# http://phylo.wikidot.com/biogeobears#texteditors
#
# 3. The PHYLIP format is the same format used for C++ LAGRANGE geography files.
#
# 4. All names in the geography file must match names in the phylogeny file.
#
# 5. DON'T USE SPACES IN SPECIES NAMES, USE E.G. "_"
#
# 6. Operational taxonomic units (OTUs) should ideally be phylogenetic lineages,
# i.e. genetically isolated populations. These may or may not be identical
# with species. You would NOT want to just use specimens, as each specimen
# automatically can only live in 1 area, which will typically favor DEC+J
# models. This is fine if the species/lineages really do live in single areas,
# but you wouldn't want to assume this without thinking about it at least.
# In summary, you should collapse multiple specimens into species/lineages if
# data indicates they are the same genetic population.
######################################################
# This is the example geography file for Hawaiian Psychotria
# (from Ree & Smith 2008)
geogfn = np(paste(addslash(extdata_dir), "Psychotria_geog.data", sep=""))
# Look at your geographic range data:
tipranges = getranges_from_LagrangePHYLIP(lgdata_fn=geogfn)
tipranges
# Maximum range size observed:
max(rowSums(dfnums_to_numeric(tipranges@df)))
# Set the maximum number of areas any species may occupy; this cannot be larger
# than the number of areas you set up, but it can be smaller.
max_range_size = 4
#######################################################
#######################################################
# DEC AND DEC+J ANALYSIS
#######################################################
#######################################################
# NOTE: The BioGeoBEARS "DEC" model is identical with
# the Lagrange DEC model, and should return identical
# ML estimates of parameters, and the same
# log-likelihoods, for the same datasets.
#
# Ancestral state probabilities at nodes will be slightly
# different, since BioGeoBEARS is reporting the
# ancestral state probabilities under the global ML
# model, and Lagrange is reporting ancestral state
# probabilities after re-optimizing the likelihood
# after fixing the state at each node. These will
# be similar, but not identical. See Matzke (2014),
# Systematic Biology, for discussion.
#
# Also see Matzke (2014) for presentation of the
# DEC+J model.
#######################################################
#######################################################
#######################################################
#######################################################
#######################################################
# Run DEC
#######################################################
# Intitialize a default model (DEC model)
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
# Give BioGeoBEARS the location of the phylogeny Newick file
BioGeoBEARS_run_object$trfn = trfn
# Give BioGeoBEARS the location of the geography text file
BioGeoBEARS_run_object$geogfn = geogfn
# Input the maximum range size
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
# 1. Here, un-comment ONLY the files you want to use.
# 2. Also un-comment "BioGeoBEARS_run_object = section_the_tree(...", below.
# 3. For example files see (a) extdata_dir,
# or (b) http://phylo.wikidot.com/biogeobears#files
# and BioGeoBEARS Google Group posts for further hints)
# Get the local copies of dispersal multipliers and times files
timesfn = np(system.file("extdata/examples/Psychotria_M0strat/BGB/timeperiods.txt", package="BioGeoBEARS"))
dispersal_multipliers_fn = np(system.file("extdata/examples/Psychotria_M0strat/BGB/dispersal_multipliers.txt", package="BioGeoBEARS"))
# Uncomment files you wish to use in time-stratified analyses:
BioGeoBEARS_run_object$timesfn = timesfn
BioGeoBEARS_run_object$dispersal_multipliers_fn = dispersal_multipliers_fn
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = FALSE # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
# (use more cores to speed it up; this requires
# library(parallel) and/or library(snow). The package "parallel"
# is now default on Macs in R 3.0+, but apparently still
# has to be typed on some Windows machines. Note: apparently
# parallel works on Mac command-line R, but not R.app.
# BioGeoBEARS checks for this and resets to 1
# core with R.app)
# Sparse matrix exponentiation is an option for huge numbers of ranges/states (600+)
# I have experimented with sparse matrix exponentiation in EXPOKIT/rexpokit,
# but the results are imprecise and so I haven't explored it further.
# In a Bayesian analysis, it might work OK, but the ML point estimates are
# not identical.
# Also, I have not implemented all functions to work with force_sparse=TRUE.
# Volunteers are welcome to work on it!!
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up DEC model
# (nothing to do; defaults)
# Look at the BioGeoBEARS_run_object; it's just a list of settings etc.
BioGeoBEARS_run_object
# This contains the model object
BioGeoBEARS_run_object$BioGeoBEARS_model_object
# This table contains the parameters of the model
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table
# Run this to check inputs. Read the error messages if you get them!
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
# For a slow analysis, run once, then set runslow=FALSE to just
# load the saved result.
runslow = TRUE
resfn = "Psychotria_DEC_M0_unconstrained_v1.Rdata"
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resDECstrat = res
} else {
# Loads to "res"
load(resfn)
resDECstrat = res
}
resDECstrat
d = resDECstrat$output@params_table["d", "est"]
e = resDECstrat$output@params_table["e", "est"]
lnL = resDECstrat$total_loglikelihood
# Python 2013 Lagrange result
# -lnL = 34.54
# dispersal = 0.03505
# extinction = 0.02831
# C++ Lagrange result
# Optimizing (simplex) -ln likelihood.
# dis: 0.0350117 ext: 0.0282904
# final -ln likelihood: 34.542
cat("\n")
txt = "Checking that BioGeoBEARS inferred DEC lnL == Lagrange DEC (Python or C++) lnL on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=lnL, digits=2), expected=round(x=-34.54196, digits=2))
cat("...PASSED ")
cat("\n")
txt = "Checking that BioGeoBEARS ML inferred 'd' DEC == Lagrange DEC (Python or C++) 'd' on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=d, digits=3), expected=round(x=0.03504546, digits=3))
cat("...PASSED ")
cat("\n")
txt = "Checking that BioGeoBEARS ML inferred 'e' DEC == Lagrange DEC (Python or C++) 'e' on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=e, digits=3), expected=round(x=0.02831871, digits=3))
cat("...PASSED ")
cat("\n\n")
txt = "CHECKS PASSED: Likelihoods and ML parameter estimates match the Lagrange (Python or C++) DEC results on the Psychotria dataset, in this case using the BioGeoBEARS time-stratification code; see the original 'out-of-the-box' example script at http://phylo.wikidot.com/biogeobears#script"
cat(txt)
cat("\n\n")
}) # END test_that
test_that(desc="Check BioGeoBEARS DEC+J ML likelihood and parameters with Hawaiian Psychotria dataset (Ree & Smith 2008).", code={
# Skip the slow tests in online checks
testthat::skip_on_cran()
testthat::skip_on_travis()
library(cladoRcpp)
#source("/drives/Dropbox/_njm/__packages/cladoRcpp_setup/cladoRcpp.R")
library(BioGeoBEARS)
library(parallel) # for detectCores
library(GenSA)
# Set your working directory for output files
# default here is your home directory ("~")
# Change this as you like
wd = np("~")
setwd(wd)
# Double-check your working directory with getwd()
getwd()
extdata_dir = np(system.file("extdata", package="BioGeoBEARS"))
extdata_dir
list.files(extdata_dir)
# This is the example Newick file for Hawaiian Psychotria
# (from Ree & Smith 2008)
# "trfn" = "tree file name"
trfn = np(paste(addslash(extdata_dir), "Psychotria_5.2.newick", sep=""))
# Look at your phylogeny:
tr = read.tree(trfn)
# This is the example geography file for Hawaiian Psychotria
# (from Ree & Smith 2008)
geogfn = np(paste(addslash(extdata_dir), "Psychotria_geog.data", sep=""))
# Look at your geographic range data:
tipranges = getranges_from_LagrangePHYLIP(lgdata_fn=geogfn)
tipranges
# Maximum range size observed:
max(rowSums(dfnums_to_numeric(tipranges@df)))
# Set the maximum number of areas any species may occupy; this cannot be larger
# than the number of areas you set up, but it can be smaller.
max_range_size = 4
#######################################################
#######################################################
#######################################################
# Run DEC+J
#######################################################
# Intitialize a default model (DEC model)
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
# Give BioGeoBEARS the location of the phylogeny Newick file
BioGeoBEARS_run_object$trfn = trfn
# Give BioGeoBEARS the location of the geography text file
BioGeoBEARS_run_object$geogfn = geogfn
# Input the maximum range size
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
# 1. Here, un-comment ONLY the files you want to use.
# 2. Also un-comment "BioGeoBEARS_run_object = section_the_tree(...", below.
# 3. For example files see (a) extdata_dir,
# or (b) http://phylo.wikidot.com/biogeobears#files
# and BioGeoBEARS Google Group posts for further hints)
#
# Uncomment files you wish to use in time-stratified analyses:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = FALSE # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
# (use more cores to speed it up; this requires
# library(parallel) and/or library(snow). The package "parallel"
# is now default on Macs in R 3.0+, but apparently still
# has to be typed on some Windows machines. Note: apparently
# parallel works on Mac command-line R, but not R.app.
# BioGeoBEARS checks for this and resets to 1
# core with R.app)
# Sparse matrix exponentiation is an option for huge numbers of ranges/states (600+)
# I have experimented with sparse matrix exponentiation in EXPOKIT/rexpokit,
# but the results are imprecise and so I haven't explored it further.
# In a Bayesian analysis, it might work OK, but the ML point estimates are
# not identical.
# Also, I have not implemented all functions to work with force_sparse=TRUE.
# Volunteers are welcome to work on it!!
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up DEC+J model
# Get the ML parameter values from the 2-parameter nested model
# (this will ensure that the 3-parameter model always does at least as good)
dstart = 0.035
estart = 0.028
jstart = 0.0001
# Input starting values for d, e
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","init"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","est"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","init"] = estart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","est"] = estart
# Add j as a free parameter
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","type"] = "free"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","init"] = jstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","est"] = jstart
# Look at the BioGeoBEARS_run_object; it's just a list of settings etc.
BioGeoBEARS_run_object
# This contains the model object
BioGeoBEARS_run_object$BioGeoBEARS_model_object
# This table contains the parameters of the model
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table
# Run this to check inputs. Read the error messages if you get them!
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
# For a slow analysis, run once, then set runslow=FALSE to just
# load the saved result.
runslow = TRUE
resfn = "Psychotria_DEC+J_M0_unconstrained_v1.Rdata"
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resDECj = res
} else {
# Loads to "res"
load(resfn)
resDECj = res
}
resDECj
# Matzke (2014) DEC+J result on Psychotria dataset
d = resDECj$output@params_table["d", "est"]
e = resDECj$output@params_table["e", "est"]
j = resDECj$output@params_table["j", "est"]
lnL = resDECj$total_loglikelihood
cat("\n")
txt = "Checking that BioGeoBEARS inferred DEC+J lnL == the original example script result on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=lnL, digits=1), expected=round(x=-20.94759, digits=1))
cat("...PASSED ")
cat("\n")
txt = "Checking that BioGeoBEARS ML inferred 'd' DEC == the original example script result on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=d, digits=3), expected=round(x=1e-12, digits=3))
cat("...PASSED ")
cat("\n")
txt = "Checking that BioGeoBEARS ML inferred 'e' DEC == the original example script result on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=e, digits=3), expected=round(x=1e-12, digits=3))
cat("...PASSED ")
cat("\n")
txt = "Checking that BioGeoBEARS ML inferred 'e' DEC == Lagrange DEC (Python or C++) 'e' on the Psychotria dataset..."
cat(txt)
expect_equal(object=round(x=j, digits=2), expected=round(x=0.1142712, digits=2))
cat("...PASSED ")
cat("\n\n")
txt = "CHECKS PASSED: Likelihoods and ML parameter estimates match the original example script for 2 basic models (DEC and DEC+J) on the Psychotria dataset, published in Matzke (2014); see the original 'out-of-the-box' example script at http://phylo.wikidot.com/biogeobears#script"
cat(txt)
cat("\n\n")
}) # END test_that
test_that(desc="check the 6 BioGeoBEARS base models on Hawaiian Psychotria", code={
# Skip the slow tests in online checks
testthat::skip_on_cran()
testthat::skip_on_travis()
library(cladoRcpp)
#source("/drives/Dropbox/_njm/__packages/cladoRcpp_setup/cladoRcpp.R")
library(BioGeoBEARS)
library(parallel) # for detectCores
library(GenSA)
# Set your working directory for output files
# default here is your home directory ("~")
# Change this as you like
wd = np("~")
setwd(wd)
# Double-check your working directory with getwd()
getwd()
# However, you can find the extdata directory like this:
extdata_dir = np(system.file("extdata", package="BioGeoBEARS"))
extdata_dir
#######################################################
# This is the example Newick file for Hawaiian Psychotria
# (from Ree & Smith 2008)
# "trfn" = "tree file name"
trfn = np(paste(addslash(extdata_dir), "Psychotria_5.2.newick", sep=""))
# Look at your phylogeny:
tr = read.tree(trfn)
# This is the example geography file for Hawaiian Psychotria
# (from Ree & Smith 2008)
geogfn = np(paste(addslash(extdata_dir), "Psychotria_geog.data", sep=""))
# Look at your geographic range data:
tipranges = getranges_from_LagrangePHYLIP(lgdata_fn=geogfn)
tipranges
# Maximum range size observed:
max(rowSums(dfnums_to_numeric(tipranges@df)))
# Set the maximum number of areas any species may occupy; this cannot be larger
# than the number of areas you set up, but it can be smaller.
max_range_size = 4
#######################################################
#######################################################
# DEC AND DEC+J ANALYSIS
#######################################################
#######################################################
#######################################################
# Run DEC
#######################################################
# Intitialize a default model (DEC model)
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
# Give BioGeoBEARS the location of the phylogeny Newick file
BioGeoBEARS_run_object$trfn = trfn
# Give BioGeoBEARS the location of the geography text file
BioGeoBEARS_run_object$geogfn = geogfn
# Input the maximum range size
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
# 1. Here, un-comment ONLY the files you want to use.
# 2. Also un-comment "BioGeoBEARS_run_object = section_the_tree(...", below.
# 3. For example files see (a) extdata_dir,
# or (b) http://phylo.wikidot.com/biogeobears#files
# and BioGeoBEARS Google Group posts for further hints)
#
# Uncomment files you wish to use in time-stratified analyses:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = "GenSA" # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
# (use more cores to speed it up; this requires
# library(parallel) and/or library(snow). The package "parallel"
# is now default on Macs in R 3.0+, but apparently still
# has to be typed on some Windows machines. Note: apparently
# parallel works on Mac command-line R, but not R.app.
# BioGeoBEARS checks for this and resets to 1
# core with R.app)
# Sparse matrix exponentiation is an option for huge numbers of ranges/states (600+)
# I have experimented with sparse matrix exponentiation in EXPOKIT/rexpokit,
# but the results are imprecise and so I haven't explored it further.
# In a Bayesian analysis, it might work OK, but the ML point estimates are
# not identical.
# Also, I have not implemented all functions to work with force_sparse=TRUE.
# Volunteers are welcome to work on it!!
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up DEC model
# (nothing to do; defaults)
# Look at the BioGeoBEARS_run_object; it's just a list of settings etc.
BioGeoBEARS_run_object
# This contains the model object
BioGeoBEARS_run_object$BioGeoBEARS_model_object
# This table contains the parameters of the model
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table
# Run this to check inputs. Read the error messages if you get them!
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
# For a slow analysis, run once, then set runslow=FALSE to just
# load the saved result.
runslow = TRUE
resfn = "Psychotria_DEC_M0_unconstrained_v1.Rdata"
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resDEC = res
} else {
# Loads to "res"
load(resfn)
resDEC = res
}
#######################################################
# Run DEC+J
#######################################################
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
BioGeoBEARS_run_object$trfn = trfn
BioGeoBEARS_run_object$geogfn = geogfn
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = "GenSA" # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up DEC+J model
# Get the ML parameter values from the 2-parameter nested model
# (this will ensure that the 3-parameter model always does at least as good)
dstart = resDEC$outputs@params_table["d","est"]
estart = resDEC$outputs@params_table["e","est"]
jstart = 0.0001
# Input starting values for d, e
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","init"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","est"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","init"] = estart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","est"] = estart
# Add j as a free parameter
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","type"] = "free"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","init"] = jstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","est"] = jstart
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
resfn = "Psychotria_DEC+J_M0_unconstrained_v1.Rdata"
runslow = TRUE
if (runslow)
{
#sourceall("/Dropbox/_njm/__packages/BioGeoBEARS_setup/")
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resDECj = res
} else {
# Loads to "res"
load(resfn)
resDECj = res
}
#######################################################
#######################################################
# DIVALIKE AND DIVALIKE+J ANALYSIS
#######################################################
#######################################################
# NOTE: The BioGeoBEARS "DIVALIKE" model is not identical with
# Ronquist (1997)'s parsimony DIVA. It is a likelihood
# interpretation of DIVA, constructed by modelling DIVA's
# processes the way DEC does, but only allowing the
# processes DIVA allows (widespread vicariance: yes; subset
# sympatry: no; see Ronquist & Sanmartin 2011, Figure 4).
#
# DIVALIKE is a likelihood interpretation of parsimony
# DIVA, and it is "like DIVA" -- similar to, but not
# identical to, parsimony DIVA.
#
# I thus now call the model "DIVALIKE", and you should also. ;-)
#######################################################
#######################################################
#######################################################
# Run DIVALIKE
#######################################################
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
BioGeoBEARS_run_object$trfn = trfn
BioGeoBEARS_run_object$geogfn = geogfn
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = "GenSA" # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up DIVALIKE model
# Remove subset-sympatry
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","est"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ysv","type"] = "2-j"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ys","type"] = "ysv*1/2"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["y","type"] = "ysv*1/2"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","type"] = "ysv*1/2"
# Allow classic, widespread vicariance; all events equiprobable
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01v","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01v","init"] = 0.5
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01v","est"] = 0.5
# No jump dispersal/founder-event speciation
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","type"] = "free"
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","init"] = 0.01
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","est"] = 0.01
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
runslow = TRUE
resfn = "Psychotria_DIVALIKE_M0_unconstrained_v1.Rdata"
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resDIVALIKE = res
} else {
# Loads to "res"
load(resfn)
resDIVALIKE = res
}
#######################################################
# Run DIVALIKE+J
#######################################################
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
BioGeoBEARS_run_object$trfn = trfn
BioGeoBEARS_run_object$geogfn = geogfn
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = "GenSA" # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up DIVALIKE+J model
# Get the ML parameter values from the 2-parameter nested model
# (this will ensure that the 3-parameter model always does at least as good)
dstart = resDIVALIKE$outputs@params_table["d","est"]
estart = resDIVALIKE$outputs@params_table["e","est"]
jstart = 0.0001
# Input starting values for d, e
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","init"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","est"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","init"] = estart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","est"] = estart
# Remove subset-sympatry
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","est"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ysv","type"] = "2-j"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ys","type"] = "ysv*1/2"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["y","type"] = "ysv*1/2"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","type"] = "ysv*1/2"
# Allow classic, widespread vicariance; all events equiprobable
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01v","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01v","init"] = 0.5
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01v","est"] = 0.5
# Add jump dispersal/founder-event speciation
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","type"] = "free"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","init"] = jstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","est"] = jstart
# Under DIVALIKE+J, the max of "j" should be 2, not 3 (as is default in DEC+J)
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","min"] = 0.00001
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","max"] = 1.99999
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
resfn = "Psychotria_DIVALIKE+J_M0_unconstrained_v1.Rdata"
runslow = TRUE
if (runslow)
{
#sourceall("/Dropbox/_njm/__packages/BioGeoBEARS_setup/")
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resDIVALIKEj = res
} else {
# Loads to "res"
load(resfn)
resDIVALIKEj = res
}
#######################################################
#######################################################
# BAYAREALIKE AND BAYAREALIKE+J ANALYSIS
#######################################################
#######################################################
# NOTE: As with DIVA, the BioGeoBEARS BayArea-like model is
# not identical with the full Bayesian model implemented
# in the "BayArea" program of Landis et al. (2013).
#
# Instead, this is a simplified likelihood interpretation
# of the model. Basically, in BayArea and BioGeoBEARS-BAYAREALIKE,
# "d" and "e" work like they do in the DEC model of Lagrange
# (and BioGeoBEARS), and then BayArea's cladogenesis assumption
# (which is that nothing in particular happens at cladogenesis) is
# replicated by BioGeoBEARS.
#
# This leaves out 3 important things that are in BayArea:
# 1. Distance dependence (you can add this with a distances
# matrix + the "x" parameter in BioGeoBEARS, however)
# 2. A correction for disallowing "e" events that drive
# a species extinct (a null geographic range)
# 3. The neat Bayesian sampling of histories, which allows
# analyses on large numbers of areas.
#
# The main purpose of having a "BAYAREALIKE" model is
# to test the importance of the cladogenesis model on
# particular datasets. Does it help or hurt the data
# likelihood if there is no special cladogenesis process?
#
# BAYAREALIKE is a likelihood interpretation of BayArea,
# and it is "like BayArea" -- similar to, but not
# identical to, Bayesian BayArea.
# I thus now call the model "BAYAREALIKE", and you should also. ;-)
#######################################################
#######################################################
#######################################################
# Run BAYAREALIKE
#######################################################
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
BioGeoBEARS_run_object$trfn = trfn
BioGeoBEARS_run_object$geogfn = geogfn
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = "GenSA" # if FALSE, use optim() instead of optimx()
BioGeoBEARS_run_object$num_cores_to_use = 1
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up BAYAREALIKE model
# No subset sympatry
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","est"] = 0.0
# No vicariance
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","est"] = 0.0
# No jump dispersal/founder-event speciation
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","type"] = "free"
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","init"] = 0.01
# BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","est"] = 0.01
# Adjust linkage between parameters
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ysv","type"] = "1-j"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ys","type"] = "ysv*1/1"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["y","type"] = "1-j"
# Only sympatric/range-copying (y) events allowed, and with
# exact copying (both descendants always the same size as the ancestor)
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","init"] = 0.9999
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","est"] = 0.9999
# Check the inputs
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
runslow = TRUE
resfn = "Psychotria_BAYAREALIKE_M0_unconstrained_v1.Rdata"
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resBAYAREALIKE = res
} else {
# Loads to "res"
load(resfn)
resBAYAREALIKE = res
}
#######################################################
# Run BAYAREALIKE+J
#######################################################
BioGeoBEARS_run_object = define_BioGeoBEARS_run()
BioGeoBEARS_run_object$trfn = trfn
BioGeoBEARS_run_object$geogfn = geogfn
BioGeoBEARS_run_object$max_range_size = max_range_size
BioGeoBEARS_run_object$min_branchlength = 0.000001 # Min to treat tip as a direct ancestor (no speciation event)
BioGeoBEARS_run_object$include_null_range = TRUE # set to FALSE for e.g. DEC* model, DEC*+J, etc.
# (For DEC* and other "*" models, please cite: Massana, Kathryn A.; Beaulieu,
# Jeremy M.; Matzke, Nicholas J.; O’Meara, Brian C. (2015). Non-null Effects of
# the Null Range in Biogeographic Models: Exploring Parameter Estimation in the
# DEC Model. bioRxiv, http://biorxiv.org/content/early/2015/09/16/026914 )
# Also: search script on "include_null_range" for other places to change
# Set up a time-stratified analysis:
#BioGeoBEARS_run_object$timesfn = "timeperiods.txt"
#BioGeoBEARS_run_object$dispersal_multipliers_fn = "manual_dispersal_multipliers.txt"
#BioGeoBEARS_run_object$areas_allowed_fn = "areas_allowed.txt"
#BioGeoBEARS_run_object$areas_adjacency_fn = "areas_adjacency.txt"
#BioGeoBEARS_run_object$distsfn = "distances_matrix.txt"
# See notes on the distances model on PhyloWiki's BioGeoBEARS updates page.
# Speed options and multicore processing if desired
BioGeoBEARS_run_object$on_NaN_error = -1e50 # returns very low lnL if parameters produce NaN error (underflow check)
BioGeoBEARS_run_object$speedup = TRUE # shorcuts to speed ML search; use FALSE if worried (e.g. >3 params)
BioGeoBEARS_run_object$use_optimx = "GenSA"
BioGeoBEARS_run_object$num_cores_to_use = 1
BioGeoBEARS_run_object$force_sparse = FALSE # force_sparse=TRUE causes pathology & isn't much faster at this scale
# This function loads the dispersal multiplier matrix etc. from the text files into the model object. Required for these to work!
# (It also runs some checks on these inputs for certain errors.)
BioGeoBEARS_run_object = readfiles_BioGeoBEARS_run(BioGeoBEARS_run_object)
# Divide the tree up by timeperiods/strata (uncomment this for stratified analysis)
#BioGeoBEARS_run_object = section_the_tree(inputs=BioGeoBEARS_run_object, make_master_table=TRUE, plot_pieces=FALSE)
# The stratified tree is described in this table:
#BioGeoBEARS_run_object$master_table
# Good default settings to get ancestral states
BioGeoBEARS_run_object$return_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_TTL_loglike_from_condlikes_table = TRUE
BioGeoBEARS_run_object$calc_ancprobs = TRUE # get ancestral states from optim run
# Set up BAYAREALIKE+J model
# Get the ML parameter values from the 2-parameter nested model
# (this will ensure that the 3-parameter model always does at least as good)
dstart = resBAYAREALIKE$outputs@params_table["d","est"]
estart = resBAYAREALIKE$outputs@params_table["e","est"]
jstart = 0.0001
# Input starting values for d, e
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","init"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","est"] = dstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","init"] = estart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","est"] = estart
# No subset sympatry
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["s","est"] = 0.0
# No vicariance
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","init"] = 0.0
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["v","est"] = 0.0
# *DO* allow jump dispersal/founder-event speciation (set the starting value close to 0)
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","type"] = "free"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","init"] = jstart
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","est"] = jstart
# Under BAYAREALIKE+J, the max of "j" should be 1, not 3 (as is default in DEC+J) or 2 (as in DIVALIKE+J)
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","max"] = 0.99999
# Adjust linkage between parameters
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ysv","type"] = "1-j"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["ys","type"] = "ysv*1/1"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["y","type"] = "1-j"
# Only sympatric/range-copying (y) events allowed, and with
# exact copying (both descendants always the same size as the ancestor)
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","type"] = "fixed"
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","init"] = 0.9999
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["mx01y","est"] = 0.9999
# NOTE (NJM, 2014-04): BAYAREALIKE+J seems to crash on some computers, usually Windows
# machines. I can't replicate this on my Mac machines, but it is almost certainly
# just some precision under-run issue, when optim/optimx tries some parameter value
# just below zero. The "min" and "max" options on each parameter are supposed to
# prevent this, but apparently optim/optimx sometimes go slightly beyond
# these limits. Anyway, if you get a crash, try raising "min" and lowering "max"
# slightly for each parameter:
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","min"] = 0.0000001
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["d","max"] = 4.9999999
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","min"] = 0.0000001
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["e","max"] = 4.9999999
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","min"] = 0.00001
BioGeoBEARS_run_object$BioGeoBEARS_model_object@params_table["j","max"] = 0.99999
check_BioGeoBEARS_run(BioGeoBEARS_run_object)
# No printing for checks
BioGeoBEARS_run_object$print_optim = FALSE
resfn = "Psychotria_BAYAREALIKE+J_M0_unconstrained_v1.Rdata"
runslow = TRUE
if (runslow)
{
res = bears_optim_run(BioGeoBEARS_run_object)
res
#save(res, file=resfn)
resBAYAREALIKEj = res
} else {
# Loads to "res"
load(resfn)
resBAYAREALIKEj = res
}
#########################################################################
#########################################################################
#########################################################################
#########################################################################
#
# CALCULATE SUMMARY STATISTICS TO COMPARE
# DEC, DEC+J, DIVALIKE, DIVALIKE+J, BAYAREALIKE, BAYAREALIKE+J
#
#########################################################################
#########################################################################
#########################################################################
#########################################################################
#########################################################################
#########################################################################
# REQUIRED READING:
#
# Practical advice / notes / basic principles on statistical model
# comparison in general, and in BioGeoBEARS:
# http://phylo.wikidot.com/advice-on-statistical-model-comparison-in-biogeobears
#########################################################################
#########################################################################
# Set up empty tables to hold the statistical results
restable = NULL
teststable = NULL
#######################################################
# Statistics -- DEC vs. DEC+J
#######################################################
# We have to extract the log-likelihood differently, depending on the
# version of optim/optimx
LnL_2 = get_LnL_from_BioGeoBEARS_results_object(resDEC)
LnL_1 = get_LnL_from_BioGeoBEARS_results_object(resDECj)
numparams1 = 3
numparams2 = 2
stats = AICstats_2models(LnL_1, LnL_2, numparams1, numparams2)
stats
# DEC, null model for Likelihood Ratio Test (LRT)
res2 = extract_params_from_BioGeoBEARS_results_object(results_object=resDEC, returnwhat="table", addl_params=c("j"), paramsstr_digits=4)
# DEC+J, alternative model for Likelihood Ratio Test (LRT)
res1 = extract_params_from_BioGeoBEARS_results_object(results_object=resDECj, returnwhat="table", addl_params=c("j"), paramsstr_digits=4)
# The null hypothesis for a Likelihood Ratio Test (LRT) is that two models
# confer the same likelihood on the data. See: Brian O'Meara's webpage:
# http://www.brianomeara.info/tutorials/aic
# ...for an intro to LRT, AIC, and AICc
rbind(res2, res1)
tmp_tests = conditional_format_table(stats)
restable = rbind(restable, res2, res1)
teststable = rbind(teststable, tmp_tests)
#######################################################
# Statistics -- DIVALIKE vs. DIVALIKE+J
#######################################################
# We have to extract the log-likelihood differently, depending on the
# version of optim/optimx
LnL_2 = get_LnL_from_BioGeoBEARS_results_object(resDIVALIKE)
LnL_1 = get_LnL_from_BioGeoBEARS_results_object(resDIVALIKEj)
numparams1 = 3
numparams2 = 2
stats = AICstats_2models(LnL_1, LnL_2, numparams1, numparams2)
stats
# DIVALIKE, null model for Likelihood Ratio Test (LRT)
res2 = extract_params_from_BioGeoBEARS_results_object(results_object=resDIVALIKE, returnwhat="table", addl_params=c("j"), paramsstr_digits=4)
# DIVALIKE+J, alternative model for Likelihood Ratio Test (LRT)
res1 = extract_params_from_BioGeoBEARS_results_object(results_object=resDIVALIKEj, returnwhat="table", addl_params=c("j"), paramsstr_digits=4)
rbind(res2, res1)
conditional_format_table(stats)
tmp_tests = conditional_format_table(stats)
restable = rbind(restable, res2, res1)
teststable = rbind(teststable, tmp_tests)
#######################################################
# Statistics -- BAYAREALIKE vs. BAYAREALIKE+J
#######################################################
# We have to extract the log-likelihood differently, depending on the
# version of optim/optimx
LnL_2 = get_LnL_from_BioGeoBEARS_results_object(resBAYAREALIKE)
LnL_1 = get_LnL_from_BioGeoBEARS_results_object(resBAYAREALIKEj)
numparams1 = 3
numparams2 = 2
stats = AICstats_2models(LnL_1, LnL_2, numparams1, numparams2)
stats
# BAYAREALIKE, null model for Likelihood Ratio Test (LRT)
res2 = extract_params_from_BioGeoBEARS_results_object(results_object=resBAYAREALIKE, returnwhat="table", addl_params=c("j"), paramsstr_digits=4)
# BAYAREALIKE+J, alternative model for Likelihood Ratio Test (LRT)
res1 = extract_params_from_BioGeoBEARS_results_object(results_object=resBAYAREALIKEj, returnwhat="table", addl_params=c("j"), paramsstr_digits=4)
rbind(res2, res1)
conditional_format_table(stats)
tmp_tests = conditional_format_table(stats)
restable = rbind(restable, res2, res1)
teststable = rbind(teststable, tmp_tests)
#########################################################################
# ASSEMBLE RESULTS TABLES: DEC, DEC+J, DIVALIKE, DIVALIKE+J, BAYAREALIKE, BAYAREALIKE+J
#########################################################################
teststable$alt = c("DEC+J", "DIVALIKE+J", "BAYAREALIKE+J")
teststable$null = c("DEC", "DIVALIKE", "BAYAREALIKE")
row.names(restable) = c("DEC", "DEC+J", "DIVALIKE", "DIVALIKE+J", "BAYAREALIKE", "BAYAREALIKE+J")
restable = put_jcol_after_ecol(restable)
restable
# Look at the results!!
restable
teststable
#######################################################
# Save the results tables for later -- check for e.g.
# convergence issues
#######################################################
#######################################################
# Model weights of all six models
#######################################################
restable2 = restable
# With AICs:
AICtable = calc_AIC_column(LnL_vals=restable$LnL, nparam_vals=restable$numparams)
restable = cbind(restable, AICtable)
restable_AIC_rellike = AkaikeWeights_on_summary_table(restable=restable, colname_to_use="AIC")
restable_AIC_rellike = put_jcol_after_ecol(restable_AIC_rellike)
restable_AIC_rellike
# With AICcs -- factors in sample size
samplesize = length(tr$tip.label)
AICtable = calc_AICc_column(LnL_vals=restable$LnL, nparam_vals=restable$numparams, samplesize=samplesize)
restable2 = cbind(restable2, AICtable)
restable_AICc_rellike = AkaikeWeights_on_summary_table(restable=restable2, colname_to_use="AICc")
restable_AICc_rellike = put_jcol_after_ecol(restable_AICc_rellike)
restable_AICc_rellike
# Check lnLs and params
lnLs = round(x=restable$LnL, digits=1)
d_inf = round(x=restable$d, digits=3)
e_inf = round(x=restable$e, digits=3)
j_inf = round(x=restable$j, digits=3)
correct_lnLs = c(-34.5, -20.9, -33.1, -21.1, -40.3, -21.6)
correct_ds = c(0.035, 0.000, 0.045, 0.000, 0.019, 0.000)
correct_es = c(0.028, 0.000, 0.000, 0.000, 0.306, 0.000)
correct_js = c(0.000, 0.114, 0.000, 0.116, 0.000, 0.108)
cat("\n")
txt = "Checking the BioGeoBEARS lnL under the 6 base models in the example script == the original example script archived results on the Psychotria dataset..."
cat(txt)
expect_equal(object=lnLs, expected=correct_lnLs)
cat("...PASSED ")
cat("\n")
txt = "Checking the BioGeoBEARS ML inferred 'd' under the 6 base models in the example script == the original example script archived results on the Psychotria dataset..."
cat(txt)
expect_equal(object=d_inf, expected=correct_ds)
cat("...PASSED ")
cat("\n")
txt = "Checking the BioGeoBEARS ML inferred 'e' under the 6 base models in the example script == the original example script archived results on the Psychotria dataset..."
cat(txt)
expect_equal(object=e_inf, expected=correct_es)
cat("...PASSED ")
cat("\n")
txt = "Checking the BioGeoBEARS ML inferred 'j' under the 6 base models in the example script == the original example script archived results on the Psychotria dataset..."
cat(txt)
expect_equal(object=j_inf, expected=correct_js)
cat("...PASSED ")
cat("\n\n")
txt = "CHECKS PASSED: Likelihoods and ML parameter estimates match the original example script for the 6 basic models on the Psychotria dataset; see the original 'out-of-the-box' example script at http://phylo.wikidot.com/biogeobears#script"
cat(txt)
cat("\n\n")
}) # END test_that
test_that(desc="check ", code={
numstates = numstates_from_numareas(numareas=4, maxareas=4, include_null_range=TRUE)
expect_equal(object=numstates, expected=16)
}) # END test_that
test_that(desc="check ", code={
numstates = numstates_from_numareas(numareas=4, maxareas=4, include_null_range=TRUE)
expect_equal(object=numstates, expected=16)
}) # END test_that
test_that(desc="check ", code={
numstates = numstates_from_numareas(numareas=4, maxareas=4, include_null_range=TRUE)
expect_equal(object=numstates, expected=16)
}) # END test_that
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.