r-cran-phylobase 0.8.2-1 / usr / lib / R / site-library / phylobase / doc / phylobase.Rnw

\documentclass{article}
%\VignetteEngine{knitr::knitr}
%\VignetteIndexEntry{phylo4: classes and methods for phylogenetic trees and data}

\usepackage[utf8]{inputenc} % for UTF-8/single quotes from sQuote()
\usepackage{graphicx}
\usepackage{array}
\usepackage{url}


%% Use a little bit more of the page
%% borrowed from Rd.sty, of r-project.org
\addtolength{\textheight}{12mm}
\addtolength{\topmargin}{-9mm}   % still fits on US paper
\addtolength{\textwidth}{24mm}   % still fits on US paper
\setlength{\oddsidemargin}{10mm}
\setlength{\evensidemargin}{\oddsidemargin}

\newcommand{\code}[1]{{{\tt #1}}}

\title{The \code{phylo4} S4 classes and methods}
\author{Ben Bolker, Peter Cowan \& Fran\c{c}ois Michonneau}
\date{\today}

\begin{document}


<<setup, include=FALSE>>=
library(knitr)
opts_chunk$set(
    fig.keep='none', dev='pdf', fig.width=6, fig.height=6,
    latex.options.color="usenames,dvipsnames"
)
@


\maketitle
\tableofcontents

\section{Introduction}

This document describes the new \code{phylo4} S4 classes and methods, which are
intended to provide a unifying standard for the representation of phylogenetic
trees and comparative data in R.  The \code{phylobase} package was developed to
help both end users and package developers by providing a common suite of tools
likely to be shared by all packages designed for phylogenetic analysis,
facilities for data and tree manipulation, and standardization of formats.

This standardization will benefit \emph{end-users} by making it easier to move
data and compare analyses across packages, and to keep comparative data
synchronized with phylogenetic trees. Users will also benefit from a repository
of functions for tree manipulation, for example tools for including or excluding
subtrees (and associated phenotypic data) or improved tree and data plotting
facilities. \code{phylobase} will benefit \emph{developers} by freeing them to
put their programming effort into developing new methods rather than into
re-coding base tools. We (the \code{phylobase} developers) hope \code{phylobase}
will also facilitate code validation by providing a repository for benchmark
tests, and more generally that it will help catalyze community development of
comparative methods in R.

A more abstract motivation for developing \code{phylobase} was to improve data
checking and abstraction of the tree data formats. \code{phylobase} can check
that data and trees are associated in the proper fashion, and protects users and
developers from accidently reordering one, but not the other. It also seeks to
abstract the data format so that commonly used information (for example, branch
length information or the ancestor of a particular node) can be accessed without
knowledge of the underlying data structure (i.e., whether the tree is stored as
a matrix, or a list, or a parenthesis-based format). This is achieved through
generic \code{phylobase} functions which which retrieve the relevant information
from the data structures. The benefits of such abstraction are multiple: (1)
\emph{easier access to the relevant information} via a simple function call
(this frees both users and developers from learning details of complex data
structures), (2) \emph{freedom to optimize data structures in the future without
  breaking code.} Having the generic functions in place to ``translate'' between
the data structures and the rest of the program code allows program and data
structure development to proceed somewhat independently. The alternative is code
written for specific data structures, in which modifications to the data
structure requires rewriting the entire package code (often exacting too high a
price, which results in the persistence of less-optimal data structures). (3)
\emph{providing broader access to the range of tools in
  \code{phylobase}}. Developers of specific packages can use these new tools
based on S4 objects without knowing the details of S4 programming.

The base \code{phylo4} class is modeled on the the \code{phylo} class in
\code{ape}.  \code{phylo4d} and \code{multiphylo4} extend the \code{phylo4}
class to include data or multiple trees respectively.  In addition to describing
the classes and methods, this vignette gives examples of how they might be used.

\section{Package overview}

The phylobase package currently implements the following functions and data structures:

\begin{itemize}
\item Data structures for storing a single tree and multiple trees:
  \code{phylo4} and \code{multiPhylo4}?
\item A data structure for storing a tree with associated tip and node data:
  \code{phylo4d}
\item A data structure for storing multiple trees with one set of tip data:
  \code{multiPhylo4d}
\item Functions for reading nexus files into the above data structures
\item Functions for converting between the above data structures and \code{ape
    phylo} objects as well as \code{ade4} \code{phylog} objects (although the
  latter are now deprecated \ldots)
\item Functions for editing trees and data (i.e., subsetting and replacing)
\item Functions for plotting trees and trees with data
\end{itemize}

\section{Using the S4 help system}

The \code{S4} help system works similarly to the \code{S3} help system with some
small differences relating to how \code{S4} methods are written.  The
\code{plot()} function is a good example.  When we type \code{?plot} we are
provided the help for the default plotting function which expects \code{x} and
\code{y}.  \code{R} also provides a way to smartly dispatch the right type of
plotting function.  In the case of an \code{ape phylo} object (a \code{S3} class
object) \code{R} evaluates the class of the object and finds the correct
functions, so the following works correctly.

<<randtree1,fig.keep='none',tidy=FALSE>>=
library(ape)
set.seed(1)  ## set random-number seed
rand_tree <- rcoal(10) ## Make a random tree with 10 tips
plot(rand_tree)
@

However, typing \code{?plot} still takes us to the default \code{plot} help.  We
have to type \code{?plot.phylo} to find what we are looking for.  This is
because \code{S3} generics are simply functions with a dot and the class name
added.

The \code{S4} generic system is too complicated to describe here, but doesn't
include the same dot notation.  As a result \code{?plot.phylo4} doesn't work,
\code{R} still finds the right plotting function.

<<convtree,fig.keep='none'>>=
library(phylobase)
# convert rand_tree to a phylo4 object
rand_p4_tree <- as(rand_tree, "phylo4")
plot(rand_p4_tree)
@

All fine and good, but how to we find out about all the great features of the
\code{phylobase} plotting function?  \code{R} has two nifty ways to find it, the
first is to simply put a question mark in front of the whole call:

<<doc0, eval=FALSE, purl=FALSE>>=
`?`(plot(rand_p4_tree))
@

\code{R} looks at the class of the \code{rand\_p4\_tree} object and takes us to
the correct help file (note: this only works with \code{S4} objects).  The
second ways is handy if you already know the class of your object, or want to
compare to generics for different classes:

<<doc1, eval=FALSE, purl=FALSE>>=
`?`(method, plot("phylo4"))
@

More information about how \code{S4} documentation works can be found in the
methods package, by running the following command.

<<doc2,eval=FALSE, purl=FALSE>>=
help('Documentation', package="methods")
@

\section{Trees without data}

You can start with a tree --- an object of class \code{phylo} from the
\code{ape} package (e.g., read in using the \code{read.tree()} or
\code{read.nexus()} functions), and convert it to a \code{phylo4} object.

For example, load the raw \emph{Geospiza} data:
<<geodata,tidy=FALSE>>=
library(phylobase)
data(geospiza_raw)
## what does it contain?
names(geospiza_raw)
@

Convert the \code{S3} tree to a \code{S4 phylo4} object using the \code{as()}
function:

<<convgeodata>>=
(g1 <- as(geospiza_raw$tree, "phylo4"))
@

The (internal) nodes appear with labels \verb+<NA>+ because
they are not defined:

<<nodelabelgeodata>>=
nodeLabels(g1)
@

You can also retrieve the node labels with \code{labels(g1,"internal")}).

A simple way to assign the node numbers as labels (useful for various checks) is

<<>>=
nodeLabels(g1) <- paste("N", nodeId(g1, "internal"), sep="")
head(g1, 5)
@

The \code{summary} method gives a little extra information, including
information on the distribution of branch lengths:

<<sumgeodata>>=
summary(g1)
@

Print tip labels:
<<tiplabelgeodata>>=
tipLabels(g1)
@

(\code{labels(g1,"tip")} would also work.)

You can modify labels and other aspects of the tree --- for example, to convert
all the labels to lower case:

<<modlabelsgeodata>>=
tipLabels(g1) <- tolower(tipLabels(g1))
@

You could also modify selected labels, e.g. to modify the labels in positions 11
and 13 (which happen to be the only labels with uppercase letters):

<<eval=FALSE, purl=FALSE>>=
tipLabels(g1)[c(11, 13)] <- c("platyspiza", "pinaroloxias")
@

Note that for a given tree, \code{phylobase} always return the \code{tipLabels}
in the same order.

Print node numbers (in edge matrix order):
<<nodenumbergeodata>>=
nodeId(g1, type='all')
@

Does it have information on branch lengths?
<<hasbrlengeodata>>=
hasEdgeLength(g1)
@

It does! What do they look like?
<<edgeLength-geodata>>=
edgeLength(g1)
@

Note that the root has \verb+<NA>+ as its length.

Print edge labels (also empty in this case --- therefore all \code{NA}):

<<edgelabelgeodata>>=
edgeLabels(g1)
@

You can also use this function to label specific edges:
<<edgelabel-assign-geodata>>=
edgeLabels(g1)["23-24"] <- "an edge"
edgeLabels(g1)
@

The edge labels are named according to the nodes they connect
(ancestor-descendant). You can get the edge(s) associated with a particular
node:

<<getEdge-geodata>>=
getEdge(g1, 24) # default uses descendant node
getEdge(g1, 24, type="ancestor") # edges using ancestor node
@

These results can in turn be passed to the function \code{edgeLength} to
retrieve the length of a given set of edges:

<<getEdge-edgeLength>>=
edgeLength(g1)[getEdge(g1, 24)]
edgeLength(g1)[getEdge(g1, 24, "ancestor")]
@

Is it rooted?

<<rootedgeodata>>=
isRooted(g1)
@

Which node is the root?
<<rootnodegeodata>>=
rootNode(g1)
@

Does it contain any polytomies?
<<polygeodata>>=
hasPoly(g1)
@

Is the tree ultrametric?
<<ultrametric-geodata>>=
isUltrametric(g1)
@

You can also get the depth (distance from the root) of any given node or the
tips:
<<nodeDepth-geodata>>=
nodeDepth(g1, 23)
depthTips(g1)
@

\section{Trees with data}

The \code{phylo4d} class matches trees with data, or combines them with a data
frame to make a \code{phylo4d} (tree-with-data) object.

Now we'll take the \emph{Geospiza} data from \verb+geospiza_raw$data+ and merge
it with the tree. First, let's prepare the data:

<<dataprep>>=
g1 <- as(geospiza_raw$tree, "phylo4")
geodata <- geospiza_raw$data
@


However, since \emph{G. olivacea} is included in the tree but
not in the data set, we will initially run into some trouble:

<<geomergedata, error=TRUE, purl=FALSE>>=
g2 <- phylo4d(g1, geodata)
@

<<echo=FALSE, results='hide'>>=
geodata <- geospiza_raw$data
@

To deal with \emph{G. olivacea} missing from the data, we have a few
choices. The easiest is to use \code{missing.data="warn"} to allow \code{R} to
create the new object with a warning (you can also use \code{missing.data="OK"}
to proceed without warnings):

<<geomerge2, tidy=FALSE, warning=TRUE, purl=FALSE>>=
g2 <- phylo4d(g1, geodata, missing.data="warn")
@

<<echo=FALSE, results='hide'>>=
g2 <- phylo4d(g1, geodata, missing.data="OK", extra.data="OK")
@

Another way to deal with this would be to use \code{prune()} to drop the
offending tip from the tree first:

<<geomerge3, results='hide'>>=
g1sub <- prune(g1, "olivacea")
g1B <- phylo4d(g1sub, geodata)
@

The difference between the two objects is that the species \emph{G. olivacea} is
still present in the tree but has no data (i.e., \verb+NA+) associated with
it. In the other case, \textit{G. olivacea} is not included in the tree
anymore. The approach you choose depends on the goal of your analysis.

You can summarize the new object with the function \code{summary}. It breaks
down the statistics about the traits based on whether it is associated with the
tips for the internal nodes:
<<geomergesum>>=
summary(g2)
@

Or use \code{tdata()} to extract the data (i.e., \code{tdata(g2)}). By default,
\code{tdata()} will retrieve tip data, but you can also get internal node data
only (\code{tdata(tree, "internal")}) or --- if the tip and node data have the
same format --- all the data combined (\code{tdata(tree, "allnode")}).

If you want to plot the data (e.g. for checking the input),
\code{plot(tdata(g2))} will create the default plot for the data --- in this
case, since it is a data frame [\textbf{this may change in future versions but
  should remain transparent}] this will be a \code{pairs} plot of the data.

\section{Subsetting}

The \code{subset} command offers a variety of ways of extracting portions of a
\code{phylo4} or \code{phylo4d} tree, keeping any tip/node data consistent.

\begin{description}
\item[tips.include]{give a vector of tips (names or numbers) to retain}
\item[tips.exclude]{give a vector of tips (names or numbers) to drop}
\item[mrca]{give a vector of node or tip names or numbers; extract the clade
    containing these taxa}
\item[node.subtree]{give a node (name or number); extract the subtree starting
    from this node}
\end{description}

Different ways to extract the \emph{fuliginosa}-\emph{scandens} clade:

<<geoextract,results='hide'>>=
subset(g2, tips.include=c("fuliginosa", "fortis", "magnirostris",
            "conirostris", "scandens"))
subset(g2, node.subtree=21)
subset(g2, mrca=c("scandens", "fortis"))
@

One could drop the clade by doing

<<geodrop, results='hide'>>=
subset(g2, tips.exclude=c("fuliginosa", "fortis", "magnirostris",
            "conirostris", "scandens"))
subset(g2, tips.exclude=names(descendants(g2, MRCA(g2, c("difficilis",
               "fortis")))))

@

% This isn't implemented yet

% Another approach is to pick the subtree graphically, by plotting the tree and
% using \code{identify}, which returns the identify of the node you click on
% with the mouse.
%
% <<geoident,eval=FALSE>>=
% plot(g1)
% n1 <- identify(g1)
% subset(g2,node.subtree=n1)
% @

\section{Tree-walking}

\code{phylobase} provides many functions that allows users to explore
relationships between nodes on a tree (tree-walking and tree traversal). Most
functions work by specifying the \code{phylo4} (or \code{phylo4d}) object as the
first argument, the node numbers/labels as the second argument (followed by some
additional arguments).

\code{getNode} allows you to find a node based on its node number or its
label. It returns a vector with node numbers as values and labels as names:

<<getnode>>=
data(geospiza)
getNode(geospiza, 10)
getNode(geospiza, "pauper")
@

If no node is specified, they are all returned, and if a node can't be found
it's returned as a \verb+NA+. It is possible to control what happens when a node
can't be found:

<<getnode2>>=
getNode(geospiza)
getNode(geospiza, 10:14)
getNode(geospiza, "melanogaster", missing="OK") # no warning
getNode(geospiza, "melanogaster", missing="warn") # warning!
@

\code{children} and \code{ancestor} give the immediate neighboring nodes:

<<children>>=
children(geospiza, 16)
ancestor(geospiza, 16)
@

while \code{descendants} and \code{ancestors} can traverse the tree up to the
tips or root respectively:

<<descendants>>=
descendants(geospiza, 16)    # by default returns only the tips
descendants(geospiza, "all") # also include the internal nodes
ancestors(geospiza, 20)
ancestors(geospiza, 20, "ALL") # uppercase ALL includes self
@

\code{siblings} returns the other node(s) associated with the same ancestor:

<<siblings>>=
siblings(geospiza, 20)
siblings(geospiza, 20, include.self=TRUE)
@

\code{MRCA} returns the most common recent ancestor for a set of tips, and
shortest path returns the nodes connecting 2 nodes:

<<mrca>>=
MRCA(geospiza, 1:6)
shortestPath(geospiza, 4, "pauper")
@

\section{multiPhylo4 classes}

\code{multiPhylo4} classes are not yet implemented but will be coming soon.

\section{Examples}

\subsection{Constructing a Brownian motion trait simulator}

This section will describe a way of constructing a simulator that generates
trait values for extant species (tips) given a tree with branch lengths,
assuming a model of Brownian motion.

We can use \code{as(tree,"phylo4vcov")} to coerce the tree into a
variance-covariance matrix form, and then use \code{mvrnorm} from the
\code{MASS} package to generate a set of multivariate normally distributed
values for the tips. (A benefit of this approach is that we can very quickly
generate a very large number of replicates.) This example illustrates a common
feature of working with \code{phylobase} --- combining tools from several
different packages to operate on phylogenetic trees with data.

We start with a randomly generated tree using \code{rcoal()} from \code{ape} to
generate the tree topology and branch lengths:

<<rtree2>>=
set.seed(1001)
tree <- as(rcoal(12), "phylo4")
@

Next we generate the phylogenetic variance-covariance matrix (by coercing the
tree to a \code{phylo4vcov} object) and pick a single set of normally
distributed traits (using \code{MASS:mvrnorm} to pick a multivariate normal
deviate with a variance-covariance matrix that matches the structure of the
tree).

<<vcvphylo>>=
vmat <- as(tree, "phylo4vcov")
vmat <- cov2cor(vmat)
library(MASS)
trvec <- mvrnorm(1, mu=rep(0, 12), Sigma=vmat)
@

The last step (easy) is to convert the \code{phylo4vcov} object back to a
\code{phylo4d} object:

<<plotvcvphylo>>=
treed <- phylo4d(tree, tip.data=as.data.frame(trvec))
plot(treed)
@

% \subsubsection{The hard way}

% <<tidy=FALSE>>=
% ## add node labels so we can match to data
% nodeLabels(tree) <- as.character(nodeId(tree, "internal"))
% ## ordering will make sure that we have ancestor value
% ## defined before descendant
% tree <- reorder(tree, "preorder")
% edgemat <- edges(tree)
% ## set aside space for values
% nodevals <- numeric(nrow(edgemat))
% ## label data in edge matrix order
% names(nodevals) <- labels(tree, "all")[nodeId(tree, "all")]
% ## variance is proportional to edge length; drop first
% ## element of edge length, which is NA
% dvals <- rnorm(nrow(edgemat) - 1, sd=edgeLength(tree)[-1]^2)
% ## indexing: ind[node number] gives position in edge matrix
% ind <- order(nodeId(tree, "all"))
% for (i in 2:nrow(edgemat)) {
%   ## value of ancestor node plus change
%   nodevals[i] <- nodevals[ind[edgemat[i, 1]]] + dvals[i - 1]
%   }
% nodevals <- data.frame(nodevals)
% treed2 <- phylo4d(tree, all.data=nodevals)
% @


% ========================================
% = Table of commands, worth the effort? =
% ========================================
% \begin{tabular}{>{\tt}ll}
% \hline
% \rm Method & Description\\
% \hline
% tdata & Retrieve tip data\\
% plot & plot tree with data if present\\
% \hline
% \end{tabular}


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% Appendices %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\appendix
\section{Definitions/slots}

This section details the internal structure of the \code{phylo4},
\code{multiphylo4} (coming soon!), \code{phylo4d}, and \code{multiphylo4d}
(coming soon!) classes. The basic building blocks of these classes are the
\code{phylo4} object and a dataframe.  The \code{phylo4} tree format is largely
similar to the one used by \code{phylo} class in the package
\code{ape}\footnote{\url{http://ape.mpl.ird.fr/}}.

We use ``edge'' for ancestor-descendant relationships in the phylogeny
(sometimes called ``branches'') and ``edge lengths'' for their lengths (``branch
lengths''). Most generally, ``nodes'' are all species in the tree; species with
descendants are ``internal nodes'' (we often refer to these just as ``nodes'',
meaning clear from context); ``tips'' are species with no descendants. The
``root node'' is the node with no ancestor (if one exists).

\subsection{phylo4}
Like \code{phylo}, the main components of
the \code{phylo4} class are:
\begin{description}
\item[edge]{a 2-column matrix of integers,
    with $N$ rows for a rooted tree or
    $N-1$ rows for an unrooted tree and
    column names \code{ancestor} and \code{descendant}.
    Each row contains information on one edge in the tree.
    See below for further constraints on the edge matrix.}
\item[edge.length]{numeric list of edge lengths
    (length $N$ (rooted) or $N-1$ (unrooted) or empty (length 0))}
\item[tip.label]{character vector of tip labels (required), with length=\# of
    tips. Tip labels need not be unique, but data-tree matching with non-unique
    labels will cause an error}
\item[node.label]{character vector of node labels, length=\# of internal nodes
    or 0 (if empty).  Node labels need not be unique, but data-tree matching
    with non-unique labels will cause an error}
\item[order]{character: ``preorder'', ``postorder'', or ``unknown'' (default),
    describing the order of rows in the edge matrix.  , ``pruningwise'' and
    ``cladewise'' are accepted for compatibility with \code{ape}}
\end{description}

The edge matrix must not contain \code{NA}s, with the exception of the root
node, which has an \code{NA} for \code{ancestor}. \code{phylobase} does not
enforce an order on the rows of the edge matrix, but it stores information on
the current ordering in the \code{@order} slot --- current allowable values are
``unknown'' (the default), ``preorder'' (equivalent to ``cladewise'' in
\code{ape}) or ``postorder'' \footnote{see
  \url{http://en.wikipedia.org/wiki/Tree_traversal} for more information on
  orderings. (\code{ape}'s ``pruningwise'' is ``bottom-up'' ordering).}.

The basic criteria for the edge matrix are similar to those of \code{ape}, as
documented it's tree
specification\footnote{\url{ape.mpl.ird.fr/misc/FormatTreeR_28July2008.pdf}}. This
is a modified version of those rules, for a tree with $n$ tips and $m$ internal
nodes:

\begin{itemize}
\item Tips (no descendants) are coded $1,\ldots, n$,
  and internal nodes ($\ge 1$ descendant)
  are coded $n + 1, \ldots , n + m$
  ($n + 1$ is the root).
  Both series are numbered with no gaps.
\item The first (ancestor)
  column has only values $> n$ (internal nodes): thus, values $\le n$
  (tips) appear only in the second (descendant) column)
\item all internal nodes [not including the root] must appear in the first
  (ancestor) column at least once [unlike \code{ape}, which nominally requires
  each internal node to have at least two descendants (although it doesn't
  absolutely prohibit them and has a \code{collapse.singles} function to get rid
  of them), \code{phylobase} does allow these ``singleton nodes'' and has a
  method \code{hasSingle} for detecting them]. Singleton nodes can be useful as
  a way of representing changes along a lineage; they are used this way in the
  \code{ouch} package.

\item the number of occurrences of a node in the first column is related to the
  nature of the node: once if it is a singleton, twice if it is dichotomous
  (i.e., of degree 3 [counting ancestor as well as descendants]), three times if
  it is trichotomous (degree 4), and so on.
\end{itemize}

\code{phylobase} does not technically prohibit reticulations (nodes or tips that
appear more than once in the descendant column), but they will probably break
most of the methods. Disconnected trees, cycles, and other exotica are not
tested for, but will certainly break the methods.

We have defined basic methods for \code{phylo4}:\code{show}, \code{print}, and a
variety of accessor functions (see help files). \code{summary} does not seem to
be terribly useful in the context of a ``raw'' tree, because there is not much
to compute.

\subsection{phylo4d}

The \code{phylo4d} class extends \code{phylo4} with data. Tip data, and
(internal) node data are stored separately, but can be retrieved together or
separately with \code{tdata(x,"tip")}, \code{tdata(x,"internal")} or
\code{tdata(x,"all")}. There is no separate slot for edge data, but these can be
stored as node data associated with the descendant node.


% \subsection{multiphylo4}

\end{document}