PhyloXML

(Difference between revisions)
Jump to: navigation, search
(Performance: Added Py2.4 info)
(Performance: new numbers, ATV correction)
Line 190: Line 190:
 
|0.01
 
|0.01
 
|0.01
 
|0.01
|0.04
+
|0.02
 
|-
 
|-
 
|bcl_2.xml
 
|bcl_2.xml
 
|
 
|
 
|105 KB
 
|105 KB
 +
|0.03
 +
|0.02
 
|0.04
 
|0.04
|0.04
 
|0.06
 
 
|-
 
|-
 
|ncbi_taxonomy_mollusca.xml
 
|ncbi_taxonomy_mollusca.xml
 
|5632
 
|5632
 
|1.5 MB
 
|1.5 MB
|0.85
+
|0.65
|0.84
+
|0.62
|0.96
+
|0.67
 
|-
 
|-
 
|tol_life_on_earth_1.xml
 
|tol_life_on_earth_1.xml
 
|57124
 
|57124
 
|46 MB
 
|46 MB
|12.79
+
|10.55
|13.09
+
|10.90
|13.36
+
|9.26
 
|-
 
|-
 
|ncbi_taxonomy_metazoa.xml
 
|ncbi_taxonomy_metazoa.xml
 
|73907
 
|73907
 
|33 MB
 
|33 MB
|18.95
+
|15.99
|19.38
+
|16.01
|12.85
+
|8.98
 
|-
 
|-
 
|ncbi_taxonomy.xml
 
|ncbi_taxonomy.xml
 
|263691
 
|263691
 
|31 MB (unindented)
 
|31 MB (unindented)
|106.98
+
|97.61
|107.77
+
|97.87
|38.91
+
|26.99
 
|}
 
|}
  
On 32-bit architectures, [http://psyco.sourceforge.net/ psyco] might improve these times significantly, at the risk of increasing memory usage. (I haven't tested it.) For comparison, the Java-based parser used in Forester and ATV (see below) reads the same files about 4 times as quickly.
+
On 32-bit architectures, [http://psyco.sourceforge.net/ psyco] might improve these times significantly, at the risk of increasing memory usage. (I haven't tested it.) For comparison, the Java-based parser used in Forester and ATV (see below) reads the same files about 3-5 times as quickly, or up to 15x for the largest file.
  
 
For Python 2.4, performance depends on which ElementTree implementation is used. Using the original pure-Python elementtree, reading/parsing takes about twice as much time for all file sizes, but writing is only significantly slower for very large files.
 
For Python 2.4, performance depends on which ElementTree implementation is used. Using the original pure-Python elementtree, reading/parsing takes about twice as much time for all file sizes, but writing is only significantly slower for very large files.

Revision as of 18:57, 17 July 2009

This module handles the parsing, generation and manipulation of files in the phyloXML format.

This code is not yet part of Biopython, and therefore the documentation has not been integrated into the Biopython Tutorial yet either.

Contents

Installation

The source code for this module currently lives on the phyloxml branch in GitHub. If you're interested in testing this code before it's been merged into Biopython, follow the instructions there to create your own fork, or just clone the phyloxml branch onto your machine.

Requirements:

  • Biopython 1.50 or newer (older may work, but hasn't been tested)
  • Python 2.4 or newer
  • ElementTree module

The XML parser used in this module is ElementTree, new to the Python standard library in Python 2.5. To use this module in Python 2.4, you'll need to install a separate package that provides the ElementTree interface. Two exist:

The module attempts to import each of these compatible ElementTree implementations until it succeeds. The given XML file handle is then parsed incrementally to instantiate an object hierarchy containing the relevant phylogenetic information.

Usage

The most useful parts of this package are available from the top level of the module:

from Bio import PhyloXML

A complete phyloXML document has a root node with the tag "phyloxml". Directly under the root is a sequence of "phylogeny" elements (phylogenetic trees), possibly followed by other arbitrary data not included in the phyloXML spec. The main structural element of these phylogenetic trees is the Clade: a tree has a clade attribute, along with other attributes, and each clade contains a series of clades (and other attributes), recursively.

The child nodes and attributes of each XML node are mapped onto classes in the PhyloXML.Tree module, keeping the names the same where possible; the XML document structure is closely mirrored in the Tree.Phyloxml objects produced by the Bio.PhyloXML package.

Parsing phyloXML files

This module provides two functions, read() and parse(). Both functions accept either a file name or an open file handle, so phyloXML data can be also loaded from compressed files, StringIO objects, and so on.

The read() function returns a single Tree.Phyloxml object representing the entire file's data. The phylogenetic trees are in the "phylogenies" attribute, and any other arbitrary data is stored in "other".

>>> phx = PhyloXML.read('phyloxml_examples.xml')
>>> print phx
Phyloxml
>>> len(phx.phylogenies)
13
>>> len(phx.other)
1
>>> print phx.other
[Other(tag='alignment', namespace='http://example.org/align')]
>>> print phx.other[0].children
[Other(tag='seq', namespace='http://www.phyloxml.org', value='acgtcgcggcccgtggaagtcctctcct'),
Other(tag='seq', namespace='http://www.phyloxml.org', value='aggtcgcggcctgtggaagtcctctcct'),
Other(tag='seq', namespace='http://www.phyloxml.org', value='taaatcgc--cccgtgg-agtccc-cct')]

If you aren't interested in the "other" data, you can use parse() to iteratively construct just the phylogenetic trees contained in the file. If there's only one tree, then the next() method on the resulting generator will return it.

>>> for tree in PhyloXML.parse('phyloxml_examples.xml'):
...     print tree.name
example from Prof. Joe Felsenstein's book "Inferring Phylogenies"
example from Prof. Joe Felsenstein's book "Inferring Phylogenies"
same example, with support of type "bootstrap"
same example, with species and sequence
same example, with gene duplication information and sequence relationships
similar example, with more detailed sequence data
network, node B is connected to TWO nodes: AB and C
...
>>> tree = PhyloXML.parse('phyloxml_examples.xml').next()
>>> tree.name
'example from Prof. Joe Felsenstein\'s book "Inferring Phylogenies"'

Writing phyloXML files

The PhyloXML.Writer module exports one function to the top level of the package: write(). It accepts a Phyloxml object (the result of read() or to_phyloxml()) and either a file name or a handle to an open file-like object. Optionally, an encoding other than UTF-8 can be specified.

>>> phx = PhyloXML.read('phyloxml_examples.xml')
>>> print phx.other
[Other(tag='alignment', namespace='http://example.org/align')]
>>> phx.other = []
>>> PhyloXML.write(phx, 'ex_no_other.xml')
>>> phx_no = PhyloXML.read('ex_no_other.xml')
>>> phx_no.other
[]

Using PhyloXML objects

Standard Python syntactic sugar is supported wherever it's reasonable.

  • str() makes a string of the object's class name and an identifier, suitable for labeling a node in generated graph
  • repr() makes a string resembling the object constructor call, such that eval(repr(obj)) will return obj for simpler PhyloXML objects, and at least partially rebuild more complex objects.
  • iter() is supported by Phyloxml and Clade objects, iterating over the contained phylogenies and sub-clades, respectively
  • len() is supported by the same objects that support iteration, with expected results

Clade objects also support extended indexing:

tree = PhyloXML.parse('example.xml').next()
assert tree.clade[0] == tree.clade.clades[0]
assert tree.clade[0,1] == tree.clade.clades[0].clades[1]

Since valid Phylogeny objects always have a single clade attribute, this style of indexing is a handy way to reach specific nodes buried deep in the tree if you happen to know exactly where they are. In the future, slicing may be supported as well to grab a range a sub-clades at once.

A couple of methods allow converting a selection to a new PhyloXML object: Phylogeny.to_phyloxml() and Clade.to_phylogeny(). A few use cases:

  • Parse a phyloXML containing multiple phylogenetic trees. Check each tree sequentially, and upon finding a tree with the desired characteristic, isolate it as a new PhyloXML object.
for tree in PhyloXML.parse('example.xml'):
    if tree.name == 'monitor lizards':
        return tree.to_phyloxml()
  • Extract a specific sub-clade and make it a separate phylogeny (and probably a new phyloXML file).
tree = PhyloXML.parse('example.xml').next()
best = None
for clade in tree.clade:
    if (clade.confidences[0].type == 'bootstrap'
            and (best is None
                or clade.confidences[0].value > best.confidences[0].value)):
        best = clade
phyloxml = best.to_phylogeny(rooted=True).to_phyloxml()
PhyloXML.write(phyloxml, 'example_best.xml')

Integrating with the rest of Biopython

Since the phyloXML specification is very detailed, no existing Biopython classes are currently used directly by the phyloXML parser. Instead, methods are available for converting between phyloXML and standard Biopython types.

The Tree.Sequence class contains methods for converting to and from Biopython SeqRecord objects. This includes the molecular sequence (mol_seq) as a Seq object, and the protein domain architecture as list of SeqFeature objects.

At some point this module should be merged into the Biopython trunk, and it would be nice to have a common interface with Bio.Nexus and Newick. Should these three modules be reorganized to extract a common Bio.TreeIO interface? Let's discuss it at some point.

Utilities

Some additional tools are located in Bio.PhyloXML.Utils.

  • dump_tags -- Print the XML tags as they are encountered in the file. For testing and debugging, mainly.
>>> PhyloXML.dump_tags('phyloxml_examples.xml')
{http://www.phyloxml.org}phyloxml
{http://www.phyloxml.org}phylogeny
{http://www.phyloxml.org}name
{http://www.phyloxml.org}description
{http://www.phyloxml.org}clade
...
  • pretty_print -- Produces a plain-text representation of the entire tree. Uses str() to display nodes by default; for the longer repr() representation, add the argument show_all=True.
>>> PhyloXML.pretty_print('phyloxml_examples.xml')
Phyloxml
	Phylogeny example from Prof. Joe Felsenstein's book "Inferring Phylogenies"
		Clade
			Clade
				Clade A
				Clade B
			Clade C
...
>>> PhyloXML.pretty_print('phyloxml_examples.xml', show_all=True)
Phyloxml()
	Phylogeny(description='phyloXML allows to use either a "branch_length"
attribute or element to indicate branch lengths.', name='example from Prof. Joe
Felsenstein's book "Inferring Phylogenies"')
		Clade()
			Clade(branch_length=0.06)
				Clade(branch_length=0.102, name='A')
				Clade(branch_length=0.23, name='B')
			Clade(branch_length=0.4, name='C')
...

Performance

This parser is meant to be able to handle large files, meaning several thousand external nodes. (Benchmarks of relevant XML parsers for Python are here.) It has been tested with files of this size; for example, the complete NCBI taxonomy parses in about 100 seconds and consumes about 1.3 GB of memory. Provided enough memory is available on the system, the writer can also rebuild phyloXML files of this size.

The read() and parse() functions process a complete file in about the same amount of CPU time. Most of the underlying code is the same, and the majority of the time is spent building Clade objects (the most common node type). For small files (smaller than ncbi_taxonomy_mollusca.xml), the write() function serializes the complete object back to an equivalent file slightly slower than the corresponding read() call; for very large files, write() finishes faster than read().

Here are some times on a 2.00GHz Intel Xeon E5405 processor (only 1 CPU core used) with 7.7GB memory, running the standard Python 2.6.2 on Ubuntu 9.04, choosing the best of 3 runs for each function:

File Ext. Nodes Size (uncompressed) Read (s) Parse (s) Write (s)
apaf.xml 38 KB 0.01 0.01 0.02
bcl_2.xml 105 KB 0.03 0.02 0.04
ncbi_taxonomy_mollusca.xml 5632 1.5 MB 0.65 0.62 0.67
tol_life_on_earth_1.xml 57124 46 MB 10.55 10.90 9.26
ncbi_taxonomy_metazoa.xml 73907 33 MB 15.99 16.01 8.98
ncbi_taxonomy.xml 263691 31 MB (unindented) 97.61 97.87 26.99

On 32-bit architectures, psyco might improve these times significantly, at the risk of increasing memory usage. (I haven't tested it.) For comparison, the Java-based parser used in Forester and ATV (see below) reads the same files about 3-5 times as quickly, or up to 15x for the largest file.

For Python 2.4, performance depends on which ElementTree implementation is used. Using the original pure-Python elementtree, reading/parsing takes about twice as much time for all file sizes, but writing is only significantly slower for very large files.

Summer of Code project

This module is being developed by Eric Talevich as a project for Google Summer of Code 2009, with NESCent as the mentoring organization and Brad Chapman as the primary mentor.

Main SoC project page: PhyloSoC:Biopython support for parsing and writing phyloXML

Other software

Christian Zmasek, one of the authors of the phyloXML specification, has released some software that uses this format:

  • Forester -- a collection of Java and Ruby libraries for working with phylogenetic data
  • Archaopteryx -- Java application for the visualization of annotated phylogenetic trees (also available in applet form)

Another list is maintained at phylosoft.org.

Personal tools
Namespaces
Variants
Actions
Navigation
Toolbox