Edit this page on GitHub

Gathering and processing biogeographical data with the BioGeography module.

Introduction

BioGeography is a module under development by Nick Matzke for a Google Summer of Code 2009 project. It is run through NESCENT’s Phyloinformatics Summer of Code 2009. See the project proposal at: Biogeographical Phylogenetics for BioPython. The mentors are Stephen Smith (primary), Brad Chapman, and David Kidd. The source code is in the Bio/Geography directory of the Geography fork of the nmatzke branch on GitHub, and you can see a timeline and other info about ongoing development of the module here. The new module is being documented on the BioPython wiki as BioGeography.

Abstract: Create a BioPython module that will enable users to automatically access and parse species locality records from online biodiversity databases; link these to user-specified phylogenies; calculate basic alpha- and beta-phylodiversity summary statistics, produce input files for input into the various inference algorithms available for inferring historical biogeography; convert output from these programs into files suitable for mapping, e.g. in Google Earth (KML files).

Summary of functions

All classes and functions have been documented with standard docstrings. Code is available at the most recent GitHub commit here: http://github.com/nmatzke/biopython/commits/Geography

Tutorial

Bio.Geography is a module for gathering and processing biogeographical data. The major motivation for the module is to assist analyses of evolutionary biogeography. A variety of inference algorithms are available for such analyses, such as DIVA and lagrange. The inputs to such programs are typically (a) a phylogeny and (b) the areas inhabited by the species at the tips of the phylogeny. A researcher who has gathered data on a particular group will likely have direct access to species location data, but many large-scale analyses may require gathering large amounts of occurrence data. Automated gathering/processing of occurrence data has a variety of other applications as well, including species mapping, niche modeling, error-checking of museum records, and monitoring range changes.

Occurrence data is derived mainly from museum collections. The major source of such data is the Global Biodiversity Information Facility (GBIF). GBIF serves occurrence data recorded by hundreds of museums worldwide. GBIF occurrence data can be searched manually, and results downloaded (see examples on GBIF website) in various formats: spreadsheet, Google Earth KML, or the XML DarwinCore format.

GBIF can also be accessed via an API. Bio.Geography can process manually downloaded DarwinCore results, or access GBIF directly.

Background: organization of Bio.Geography

It is useful to understand the overall organization of classes in Bio.Geography. There are four classes within the GbifXml module:

Parsing a local (manually downloaded) GBIF DarwinCore XML file

For one-off uses of GBIF, you may find it easiest to just download occurrence data in spreadsheet format (for analysis) or KML (for mapping). But for analyses of many groups, or for repeatedly updating an analysis as new data is added to GBIF, automation is desirable.

A manual search conducted on the GBIF website can return results in the form of an XML file adhering to the DarwinCore data standard. An example file can be found in Biopython’s Tests/Geography directory, with the name utric_search_v2.xml. This file contains over 1000 occurrence records for Utricularia, a genus of carnivorous plant.

Save the utric_search_v2.xml file in your working directory (or download a similar file from GBIF). Here are suggested steps to parse the file with Bio.Geography’s GbifXml module. First, import the necessary classes and functions, and specify the filename of the input file.

from Bio.Geography.GbifXml import GbifXmlTree, GbifSearchResults
from Bio.Geography.GeneralUtils import fix_ASCII_file

xml_fn = 'utric_search_v2.xml'

Second, in order to display results to screen in python, we need to convert the file to plain ASCII (GBIF results contain all many of unusual characters from different languages, and no standardization of slanted quotes and the like; this can cause crashes when attempting to print to screen in python or ipython).

xml_fn_new = fix_ASCII_file(xml_fn)

This creates a new file with the string “_fixed.xml” added to the filename.

Next, we will parse the XML file into an ElementTree (a python object which contains the data from the XML file as a nested series of lists and dictionaries).

from xml.etree import ElementTree as ET
xmltree = ET.parse(xml_fn_new)

We can then store the element tree as an object of Class GbifXmlTree:

gbif_recs_xmltree = GbifXmlTree(xmltree)

Then, with the xmltree stored, we parse it into individual records (stored in individual objects of class GbifObservationRecord), which are then stored as a group in an object of class GbifSearchResults.

recs = GbifSearchResults(gbif_recs_xmltree)
recs.extract_occurrences_from_gbif_xmltree(recs.gbif_recs_xmltree)

The list of individual observation records can be accessed at recs.obs_recs_list. This will display the references to the first five records:

recs.obs_recs_list[0:4]

To get the data for the first individual record:

rec = recs.obs_recs_list[0]
dir(rec)

rec.lat will return the latitude, rec.long the longitude, etc. Certain data attributes are not found in all GBIF records; if they are missing, the field in question will contain “None”.

To print all of the records in a tab-delimited table format:

recs.print_records()

Checking how many matching records are hosted by GBIF

Before we go through the trouble of downloading thousands of records, we may wish to know how many there are in GBIF first. The user must set up a dictionary containing the fields and search terms as keys and items, respectively. I.e.,

from GbifXml import GbifXmlTree, GbifSearchResults

params = {'format': 'darwin', 'scientificname': 'Genlisea*'}

“‘format’: ‘darwin’” specifies that GBIF should return the results in DarwinCore format.

‘scientificname’ specifies the genus name to search on. Adding an ‘*’ after the name will return anything that begins with “Genlisea”.

The full list of search terms can be found on GBIF’s Occurrence record data service, which is linked from the Using data from the GBIF portal.

Once you have specified your search parameters, initiate a new GbifSearchResults object and run get_numhits to get the number of hits:

params = {'format': 'darwin', 'scientificname': 'Genlisea*'}
recs = GbifSearchResults()
numhits = recs.get_numhits(params)

As of August 2009, 169 matching records existed in GBIF matching “Genlisea*”

For constrast, run the same search without the asterisk (‘*’):

params = {'format': 'darwin', 'scientificname': 'Genlisea'}
numhits = recs.get_numhits(params)

We only get ~10 results – presumably records of specimens only identified down to genus and no further.

Downloading an individual record

Individual records can be downloaded by key. To download an individual record:

rec = recs.obs_recs_list[0]
key = rec.gbifkey
# (or manually)
# key = 175067484
xmlrec = recs.get_record(key)
print(xmlrec)

If you want to print the xmlrec ElementTree object, store xmlrec in a GbifXmlTree object and run print_xmltree:

GbifXmlTree(xmlrec).print_xmltree()

Summary statistics for phylogenetic trees with TreeSum

Biogeographical regions are often characterized by alpha and beta-diversity statistics: basically, these are indices of the number of species found within or between regions. Given a phylogeny for organisms in a region, phylogenetic alpha- and beta-diversity statistics can be calculated. This has been implemented in a thorough way in the phylocom package by Webb et al., but for some purposes it is useful to calculate the statistics directly in python.

Here, we need to start with a Newick tree string:

trstr2 = "(((t9:0.385832, (t8:0.445135,t4:0.41401)C:0.024032)B:0.041436, t6:0.392496)A:0.0291131, t2:0.497673, ((t0:0.301171, t7:0.482152)E:0.0268148, ((t5:0.0984167,t3:0.488578)G:0.0349662, t1:0.130208)F:0.0318288)D:0.0273876);"

to2 = Tree(trstr2)

Then, we create a tree summary object:

ts = TreeSum(to2)

The function test_Tree will run the metrics (MPD = Mean Phylogenetic Distance, NRI = Net Relatedness Index, MNPD = Mean Nearest Neighbor Phylogenetic Distance, NTI = Nearest Taxon Index, PD = total Phylogenetic distance) and output to screen:

ts.test_Tree()

By subsetting a tree to taxa only existing within a region, statistics can be calculated by region.

Downloading and processing large numbers of records

GBIF only allows a maximum of 1000 observation records to be downloaded at a time (10,000 for KML records). To get more, we need to download and process them in stages.

Again we will set up our parameters dictionary, and also an “inc” variable to specify the number of records to download per server request.

params = {'format': 'darwin', 'scientificname': 'Genlisea*'}
inc = 100
recs3 = GbifSearchResults()
gbif_xmltree_list = recs3.get_all_records_by_increment(params, inc)

As with biopython’s interactions with NCBI servers, the GbifSearchResults module keeps track of when the last GBIF request was made, and requires a 3-second wait before a new request.

Each server request returns an XML string; these are parsed into GbifXmlTree objects, and a list of the returned GbifXmlTree objects is returned to gbif_xmltree_list. The individual records have also been parsed:

recs3.print_records()

Classifying records into geographical regions

Biogeographical analyses will often require that you determine what area(s) a taxon lives in. Areas are not always obviously delineated, and analysts may wish to try several different possible sets of areas and see how this influences their analysis.

Below, we set up a polygon containing the latitude/longitude coordinates for the Northern Hemisphere, and then set the “area” attribute for each matching record to “NorthernHemisphere”:

ul = (-180, 90)
ur = (180, 90)
ll = (-180, 0)
lr = (180, 0)
poly = [ul, ur, ll, lr]
polyname = "NorthernHemisphere"

recs3.print_records()

This process can be repeated for all polygons of interest until all GBIF records have been classified (except for GBIF records which lacked lat/long data in the first place, which sometimes happens).

GeogUtils also contains open access libraries for processing shapefile/dbf files – these are standard GIS file formats, and various publicly-accessible shapefiles might serve as sources for polygons.

Warning: the point-in-polygon operation will fail dramatically if your polygon crosses the International Dateline. The best solution in this case is to split any polygons crossing the dateline into two polygons, one on each side of the line.

General notes

GBIF search results often contain non-ASCII characters (e.g. international placenames) and other confusing items, e.g., web links in angle brackets, which can be misinterpreted as unmatched XML tags if a GBIF search result is read to ASCII and then an attempt is made to parse it.

In general, the Geography module will handle things fine if the results are being processed in the background; but to print results to screen, a series of functions from GeneralUtils are used to convert a string to plain ASCII. This avoids crashes e.g. when printing data to screen. Therefore, these printed-to-screen results may slightly alter the content of the original search results.