Introduction
These notes should enable the user to estimate phylogenetic trees from alignment data with different methods using the phangorn package (Schliep 2011) . Several functions of package are also described in more detail in (Paradis 2012). For more theoretical background on all the methods see e.g. (Felsenstein 2004)(Yang 2006). This document illustrates some of the package features to estimate phylogenetic trees using different reconstruction methods. Small adaptations to the scripts in section @ref(appendix) should enable the user to perform phylogenetic analyses.
Getting started
The first thing we have to do is to read in an alignment. Unfortunately there exists many different file formats that alignments can be stored in. The function read.phyDat is used to read in an alignment. There are several functions to read in alignments depending on the format of the data set (“nexus,” “phylip,” “fasta”) and the kind of data (amino acid or nucleotides) in the ape package (Paradis and Schliep 2019) and phangorn. The function read.phyDat calls these other functions and transform them into a phyDat object. For the specific parameter settings available look in the help files of the function read.dna (for phylip, fasta, clustal format), read.nexus.data for nexus files. For amino acid data additional read.aa is called.
We start our analysis loading the phangorn package and then reading in an alignment.
library(ape)
library(phangorn)
fdir <- system.file("extdata/trees", package = "phangorn")
primates <- read.phyDat(file.path(fdir, "primates.dna"),
format = "interleaved")Distance based methods
After reading in the alignment we can build a first tree with distance based methods. The function dist.dna from the ape package computes distances for many DNA substitution models. To use the function dist.dna we have to transform the data to class DNAbin. For amino acids the function dist.ml offers common substitution models (for example “WAG,” “JTT,” “LG,” “Dayhoff,” “cpREV,” “mtmam,” “mtArt,” “MtZoa” or “mtREV24”).
After constructing a distance matrix we reconstruct a rooted tree with UPGMA and alternatively an unrooted tree using Neighbor Joining (Saitou and Nei 1987)(Studier and Keppler 1988). More distance methods like fastme are available in the ape package.
dm <- dist.ml(primates)
treeUPGMA <- upgma(dm)
treeNJ <- NJ(dm)We can plot the trees treeUPGMA and treeNJ with the commands:
plot(treeUPGMA, main="UPGMA")
Rooted UPGMA tree.
plot(treeNJ, "unrooted", main="NJ")
Unrooted NJ tree.
Bootstrap
To run the bootstrap we need to first write a function which computes a tree from an alignment. So we first need to compute a distance matrix and afterwards compute the tree. This function we can than give to the bootstrap.phyDat function.
fun <- function(x) upgma(dist.ml(x))
bs_upgma <- bootstrap.phyDat(primates, fun)With the new syntax of R 4.1 this can be written a bit shorter:
bs_upgma <- bootstrap.phyDat(primates, \(x){dist.ml(x) |> upgma})Finally we can plot the tree with bootstrap values added:
plotBS(treeUPGMA, bs_upgma, main="UPGMA")
Rooted UPGMA tree.
Distance based methods are very fast and we will use the UPGMA and NJ tree as starting trees for the maximum parsimony and maximum likelihood analyses.
Parsimony
The function parsimony returns the parsimony score, that is the number of changes which are at least necessary to describe the data for a given tree. We can compare the parsimony score or the two trees we computed so far:
parsimony(treeUPGMA, primates)## [1] 751parsimony(treeNJ, primates)## [1] 746The function optim.parsimony performs tree rearrangements to find trees with a lower parsimony score. The tree rearrangement implemented are nearest-neighbor interchanges (NNI) and subtree pruning and regrafting (SPR). The later one only works so far with the fitch algorithm.
treePars <- optim.parsimony(treeUPGMA, primates)## Final p-score 746 after 1 nni operationsHowever is also a version of the parsimony ratchet (Nixon 1999) implemented, which is likely to find better trees than just doing NNI / SPR rearrangements. The current implementation
- Create a bootstrap data set \(D_b\) from the original data set.
- Take the current best tree and perform tree rearrangements on \(D_b\) and save bootstrap tree as \(T_b\).
- Use \(T_b\) and perform tree rearrangements on the original data set. If this trees has a lower parsimony score than the currently best tree replace it.
- Iterate 1:3 until either a given number of iteration is reached or no improvements have been recorded for a number of iterations.
treeRatchet <- pratchet(primates, trace = 0)
parsimony(c(treePars, treeRatchet), primates)## [1] 746 746Next we assign branch length to the tree. The branch length are proportional to the number of substitutions / site.
treeRatchet <- acctran(treeRatchet, primates)The parsimony ratchet implicitly performs a bootstrap analysis (step 1). We make use of this and store the trees which where visited. This allows us to add bootstrap support values to the tree.
plotBS(midpoint(treeRatchet), type="phylogram")
Branch and bound
For data sets with few species it is also possible to find all most parsimonious trees using a branch and bound algorithm (Hendy and Penny 1982). For data sets with more than 10 taxa this can take a long time and depends strongly on how tree like the data are. And for more than 20-30 taxa this will take almost forever.
(trees <- bab(subset(primates,1:10)))## 1 phylogenetic treeMaximum likelihood
The last method we will describe in this vignette is Maximum Likelihood (ML) as introduced by Felsenstein (Felsenstein 1981). We can easily compute the likelihood for a tree given the data
fit = pml(treeNJ, data=primates)
fit##
## loglikelihood: -3075
##
## unconstrained loglikelihood: -1230
##
## Rate matrix:
## a c g t
## a 0 1 1 1
## c 1 0 1 1
## g 1 1 0 1
## t 1 1 1 0
##
## Base frequencies:
## 0.25 0.25 0.25 0.25The function pml returns an object of class pml. This object contains the data, the tree and many different parameters of the model like the likelihood. There are many generic functions for the class pml available, which allow the handling of these objects.
methods(class="pml")## [1] AICc BIC anova logLik plot print simSeq update vcov
## see '?methods' for accessing help and source codeThe object fit just estimated the likelihood for the tree it got supplied, but the branch length are not optimized for the Jukes-Cantor model yet, which can be done with the function optim.pml.
fitJC <- optim.pml(fit, TRUE)## optimize edge weights: -3075 --> -3068
## optimize edge weights: -3068 --> -3068
## optimize topology: -3068 --> -3068
## optimize topology: -3068 --> -3068
## NNI moves: 1
## optimize edge weights: -3068 --> -3068
## optimize topology: -3068 --> -3068
## NNI moves: 0logLik(fitJC)## 'log Lik.' -3068 (df=25)With the default valuespml will estimate a Jukes-Cantor model. The generic function update allows to change parameters. We will change the model to the GTR + \(\Gamma(4)\) + I model and then optimize all the parameters.
fitGTR <- update(fit, k=4, inv=0.2)
fitGTR <- optim.pml(fitGTR, model="GTR", optInv=TRUE, optGamma=TRUE,
rearrangement = "NNI", control = pml.control(trace = 0))
fitGTR##
## loglikelihood: -2609
##
## unconstrained loglikelihood: -1230
## Proportion of invariant sites: 0.005674
## Discrete gamma model
## Number of rate categories: 4
## Shape parameter: 2.989
##
## Rate matrix:
## a c g t
## a 0.0000 0.606984 35.533580 0.3694
## c 0.6070 0.000000 0.003543 14.9219
## g 35.5336 0.003543 0.000000 1.0000
## t 0.3694 14.921882 1.000000 0.0000
##
## Base frequencies:
## 0.3934 0.3792 0.04025 0.1872With the control parameters the thresholds for the fitting process can be changed. Here we want just to suppress output during the fitting process. For larger trees the NNI rearrangements often get stuck in a local maximum. We added two stochastic algorithms to improve topology search. The first (set rearrangement="stochastic") performs stochastic rearrangements similar as in (Nguyen et al. 2015), which makes random NNI permutation to the tree, which than gets optimized to escape local optima. The second option (rearrangement=“ratchet”) perform the likelihood ratchet (Vos 2003).
While these algorithms may find better trees they will also take more time.
fitGTR <- optim.pml(fitGTR, model="GTR", optInv=TRUE, optGamma=TRUE,
rearrangement = "stochastic", control = pml.control(trace = 0))
fitGTR##
## loglikelihood: -2607
##
## unconstrained loglikelihood: -1230
## Proportion of invariant sites: 0.005632
## Discrete gamma model
## Number of rate categories: 4
## Shape parameter: 2.701
##
## Rate matrix:
## a c g t
## a 0.0000 0.798458 74.041130 0.6375
## c 0.7985 0.000000 0.003006 28.5984
## g 74.0411 0.003006 0.000000 1.0000
## t 0.6375 28.598448 1.000000 0.0000
##
## Base frequencies:
## 0.3964 0.3788 0.04029 0.1845Model selection
We can compare nested models for the JC and GTR + \(\Gamma(4)\) + I model using likelihood ratio statistic
anova(fitJC, fitGTR)## Likelihood Ratio Test Table
## Log lik. Df Df change Diff log lik. Pr(>|Chi|)
## 1 -3068 25
## 2 -2607 35 10 923 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1with the Shimodaira-Hasegawa test
SH.test(fitGTR, fitJC)## Trees ln L Diff ln L p-value
## [1,] 1 -2607 0.0 0.5039
## [2,] 2 -3068 461.5 0.0000or with the AIC
AIC(fitJC)## [1] 6187AIC(fitGTR)## [1] 5284AICc(fitGTR)## [1] 5296BIC(fitGTR)## [1] 5404An alternative is to use the function modelTest to compare different nucleotide or protein models the AIC, AICc or BIC, similar to popular programs ModelTest and ProtTest (D. Posada and Crandall 1998), (David Posada 2008), (Abascal, Zardoya, and Posada 2005).
mt = modelTest(primates)The results of modelTest is illustrated in following table:
| Model | df | logLik | AIC | AICw | AICc | AICcw | BIC |
|---|---|---|---|---|---|---|---|
| JC | 25 | -3068 | 6187 | 0.00 | 6193 | 0.00 | 6273 |
| JC+I | 26 | -3063 | 6177 | 0.00 | 6184 | 0.00 | 6267 |
| JC+G | 26 | -3067 | 6186 | 0.00 | 6193 | 0.00 | 6275 |
| JC+G+I | 27 | -3063 | 6179 | 0.00 | 6187 | 0.00 | 6272 |
| F81 | 28 | -2918 | 5892 | 0.00 | 5900 | 0.00 | 5989 |
| F81+I | 29 | -2909 | 5876 | 0.00 | 5885 | 0.00 | 5976 |
| F81+G | 29 | -2913 | 5883 | 0.00 | 5892 | 0.00 | 5983 |
| F81+G+I | 30 | -2909 | 5877 | 0.00 | 5886 | 0.00 | 5980 |
| K80 | 26 | -2953 | 5958 | 0.00 | 5965 | 0.00 | 6048 |
| K80+I | 27 | -2945 | 5943 | 0.00 | 5950 | 0.00 | 6036 |
| K80+G | 27 | -2945 | 5944 | 0.00 | 5951 | 0.00 | 6037 |
| K80+G+I | 28 | -2942 | 5941 | 0.00 | 5949 | 0.00 | 6037 |
| HKY | 29 | -2625 | 5308 | 0.00 | 5317 | 0.00 | 5408 |
| HKY+I | 30 | -2621 | 5303 | 0.00 | 5312 | 0.00 | 5406 |
| HKY+G | 30 | -2613 | 5285 | 0.18 | 5295 | 0.45 | 5389 |
| HKY+G+I | 31 | -2612 | 5287 | 0.08 | 5297 | 0.14 | 5394 |
| SYM | 30 | -2814 | 5688 | 0.00 | 5697 | 0.00 | 5791 |
| SYM+I | 31 | -2812 | 5685 | 0.00 | 5695 | 0.00 | 5792 |
| SYM+G | 31 | -2805 | 5671 | 0.00 | 5681 | 0.00 | 5778 |
| SYM+G+I | 32 | -2805 | 5673 | 0.00 | 5684 | 0.00 | 5784 |
| GTR | 33 | -2619 | 5303 | 0.00 | 5315 | 0.00 | 5417 |
| GTR+I | 34 | -2614 | 5295 | 0.00 | 5307 | 0.00 | 5412 |
| GTR+G | 34 | -2608 | 5283 | 0.47 | 5295 | 0.29 | 5401 |
| GTR+G+I | 35 | -2607 | 5284 | 0.27 | 5297 | 0.11 | 5405 |
The thresholds for the optimization in modelTest are not as strict as for optim.pml and no tree rearrangements are performed. As modelTest computes and optimizes a lot of models it would be a waste of computer time not to save these results. The results are saved as call together with the optimized trees in an environment and this call can be evaluated to get a pml object back to use for further optimization or analysis.
env <- attr(mt, "env")
ls(envir=env)## [1] "F81" "F81+G" "F81+G+I" "F81+I" "GTR"
## [6] "GTR+G" "GTR+G+I" "GTR+I" "HKY" "HKY+G"
## [11] "HKY+G+I" "HKY+I" "JC" "JC+G" "JC+G+I"
## [16] "JC+I" "K80" "K80+G" "K80+G+I" "K80+I"
## [21] "SYM" "SYM+G" "SYM+G+I" "SYM+I" "data"
## [26] "tree_F81" "tree_F81+G" "tree_F81+G+I" "tree_F81+I" "tree_GTR"
## [31] "tree_GTR+G" "tree_GTR+G+I" "tree_GTR+I" "tree_HKY" "tree_HKY+G"
## [36] "tree_HKY+G+I" "tree_HKY+I" "tree_JC" "tree_JC+G" "tree_JC+G+I"
## [41] "tree_JC+I" "tree_K80" "tree_K80+G" "tree_K80+G+I" "tree_K80+I"
## [46] "tree_SYM" "tree_SYM+G" "tree_SYM+G+I" "tree_SYM+I"(fit <- eval(get("HKY+G+I", env), env))##
## loglikelihood: -2612
##
## unconstrained loglikelihood: -1230
## Proportion of invariant sites: 0.002694
## Discrete gamma model
## Number of rate categories: 4
## Shape parameter: 2.124
##
## Rate matrix:
## a c g t
## a 0.00 1.00 56.02 1.00
## c 1.00 0.00 1.00 56.02
## g 56.02 1.00 0.00 1.00
## t 1.00 56.02 1.00 0.00
##
## Base frequencies:
## 0.4205 0.3622 0.0439 0.1734Bootstrap
At last we may want to apply bootstrap to test how well the edges of the tree are supported:
bs = bootstrap.pml(fitJC, bs=100, optNni=TRUE,
control = pml.control(trace = 0))Now we can plot the tree with the bootstrap support values on the edges and also look at consensusNet to identify potential conflict.
plotBS(midpoint(fitJC$tree), bs, p = 50, type="p")
Tree with bootstrap support. Unrooted tree (midpoint rooted) with bootstrap support values.
cnet <- consensusNet(bs, p=0.2)
plot(cnet, show.edge.label=TRUE)
ConsensusNet from the bootstrap sample.
Molecular dating with a strict clock
We implemented a strict clock as described in (Felsenstein 2004), p. ???.
So far the starting tree needs to be ultrametric, so we use the UPGMA tree. When we optimize the tree with optim.pml we have to make sure set we set optRooted = TRUE!
fit_strict <- pml(treeUPGMA, data=primates, k=4, bf=baseFreq(primates))
fit_strict <- optim.pml(fit_strict, model="GTR", optRooted = TRUE,
rearrangement = "NNI", optGamma = TRUE, optInv = TRUE,
control = pml.control(trace = 0))plot(fit_strict)
Several analyses, e.g.bootstrap and modelTest, can be computationally demanding, but as nowadays most computers have several cores one can distribute the computations using the parallel package. However it is only possible to use this approach if R is running from command line (“X11”), but not using a GUI (for example “Aqua” on Macs) and unfortunately the parallel package does not work at all under Windows.
Appendix
Standard scripts for nucleotide analysis
Here we provide two standard scripts which can be adapted for the most common tasks. Most likely the arguments for read.phyDat have to be adapted to accommodate your file format. Both scripts assume that the parallel package works on your platform, see comments above.
library(phangorn)
file <- "myfile"
dat <- read.phyDat(file)
dm <- dist.ml(dat, "F81")
tree <- NJ(dm)
# as alternative for a starting tree:
tree <- pratchet(dat) # parsimony tree
tree <- nnls.phylo(tree, dm) # need edge weights
# 1. alternative: quick and dirty: GTR + G
fitStart <- pml(tree, dat, k=4)
fit <- optim.pml(fitStart, model="GTR", optGamma=TRUE, rearrangement="stochastic")
# 2. alternative: prepare with modelTest
mt <- modelTest(dat, tree=tree, multicore=TRUE)
mt[order(mt$AICc),]
# choose best model from the table according to AICc
bestmodel <- mt$Model[which.min(mt$AICc)]
env <- attr(mt, "env")
fitStart <- eval(get("GTR+G+I", env), env)
# or let R search the table
fitStart <- eval(get(bestmodel, env), env)
# equivalent to: fitStart = eval(get("GTR+G+I", env), env)
fit <- optim.pml(fitStart, rearrangement = "stochastic",
optGamma=TRUE, optInv=TRUE, model="GTR")
bs <- bootstrap.pml(fit, bs=100, optNni=TRUE, multicore=TRUE)Standard scripts for amino acid analysis
You can specify different several models build in which you can specify, e.g. “WAG,” “JTT,” “Dayhoff,” “LG.” Optimizing the rate matrix for amino acids is possible, but would take a long, a very long time and you will need to have a large alignment to estimate all the parameters. So make sure to set optBf=FALSE and optQ=FALSE in the function optim.pml, which is also the default.
library(phangorn)
file <- "myfile"
dat <- read.phyDat(file, type = "AA")
dm <- dist.ml(dat, model="JTT")
tree <- NJ(dm)
# parallel will only work safely from command line
# and not at all windows
(mt <- modelTest(dat, model=c("JTT", "LG", "WAG"),
multicore=TRUE))
# run all available amino acid models
(mt <- modelTest(dat, model="all", multicore=TRUE))
env <- attr(mt, "env")
fitStart <- eval(get(mt$Model[which.min(mt$BIC)], env), env)
fitNJ <- pml(tree, dat, model="JTT", k=4, inv=.2)
fit <- optim.pml(fitNJ, rearrangement = "stochastic",
optInv=TRUE, optGamma=TRUE)
fit
bs <- bootstrap.pml(fit, bs=100, optNni=TRUE, multicore=TRUE)Session info
## R version 4.1.2 (2021-11-01)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.3 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/liblapack.so.3
##
## locale:
## [1] LC_CTYPE=de_AT.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=de_AT.UTF-8 LC_COLLATE=C
## [5] LC_MONETARY=de_AT.UTF-8 LC_MESSAGES=de_AT.UTF-8
## [7] LC_PAPER=de_AT.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=de_AT.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] knitr_1.36 phangorn_2.8.1 ape_5.5-3
##
## loaded via a namespace (and not attached):
## [1] igraph_1.2.9 Rcpp_1.0.7 magrittr_2.0.1 lattice_0.20-45
## [5] R6_2.5.1 quadprog_1.5-8 rlang_0.4.12 fastmatch_1.1-3
## [9] fastmap_1.1.0 highr_0.9 stringr_1.4.0 tools_4.1.2
## [13] parallel_4.1.2 grid_4.1.2 nlme_3.1-153 xfun_0.28
## [17] jquerylib_0.1.4 htmltools_0.5.2 yaml_2.2.1 digest_0.6.29
## [21] Matrix_1.3-4 codetools_0.2-18 sass_0.4.0 prettydoc_0.4.1
## [25] evaluate_0.14 rmarkdown_2.11 stringi_1.7.6 compiler_4.1.2
## [29] bslib_0.3.1 jsonlite_1.7.2 pkgconfig_2.0.3