BIOINFORMATICS REVIEW Vol. 25 no. 4 2009, pages 421–428doi:10.1093/bioinformatics/btn659
Data and text mining
Biodiversity informatics: automated approaches for documenting
global biodiversity patterns and processes
Robert Guralnick1,2,∗ and Andrew Hill2
1University of Colorado Museum of Natural History and 2Department of Ecology and Evolutionary Biology,
University of Colorado Boulder, Boulder, CO 80309-0265, USA
Received on November 23, 2008; revised on December 16, 2008; accepted on December 18, 2008
Advance Access publication January 6, 2009
Associate Editor: Jonathan Wren
ABSTRACT
Motivation: Data about biodiversity have been scattered in different
formats in natural history collections, survey reports and the
literature. A central challenge for the biodiversity informatics
community is to provide the means to share and rapidly synthesize
these data and the knowledge they provide us to build an easily
accessible, unified global map of biodiversity. Such a map would
provide raw and summary data and information on biodiversity and
its change across the world at multiple scales.
Results: We discuss a series of steps required to create a
unified global map of biodiversity. These steps include: building
biodiversity repositories; creating scalable species distribution
maps; creating flexible, user-programmable pipelines which enable
biodiversity assessment; and integrating phylogenetic approaches
into biodiversity assessment. We show two case studies that
combine phyloinformatic and biodiversity informatic approaches to
document large scale biodiversity patterns. The first case study uses
data available from the Barcode of Life initiative in order to make
species conservation assessment of North American birds taking into
account evolutionary uniqueness. The second case study uses full
genomes of influenza A available from Genbank to provide an auto-
updating documentation of the evolution and geographic spread of
these viruses.
Availability: Both the website for tracking evolution and spread of
influenza A and the website for applying phyloinformatics analysis
to Barcode of Life data are available as outcomes of case studies
(http://biodiversity.colorado.edu).
Contact: robert.guralnick@colorado.edu
1 INTRODUCTION
Biodiversity is in crisis (Heywood and Watson, 1995; Jenkins,
2003; Loreau et al., 2006; Pimm et al., 1995; Wilson, 1998).
With the predicted loss of genetic and species diversity as great
as past mass extinction events (Novacek and Cleland, 2001), a
pressing challenge in environmental sciences will be to understand
the factors causing this decline (National Academy of Sciences,
2001). One of the first steps towards meeting this challenge will
be to document global patterns of biodiversity and note how those
patterns change over time and space. Such an endeavor will result in
∗To whom correspondence should be addressed.
what Wilson (2000) called the global biodiversity map, stretching
from pole to pole and including the tens of millions of species,
their genes, proteins, behavior and morphology. This daunting task
might seem impossible to achieve in time to make a difference
given the pace and scope of biodiversity loss. However, many
points on the map are nascent, already existing within the body of
biodiversity literature and natural history collections, representing
different views of past and present biodiversity. Thus, a central
challenge for biodiversity informaticians is to provide the means
to share and rapidly synthesize these data and the knowledge they
provide us to build a unified global map of biodiversity (Bisby, 2000;
Canhos et al., 2004).
We are in a renaissance with regards to mapping natural and
man-made environments, aided in part by remote sensing data and
new, easy to use computer-aided mapping tools that show these
data (Butler, 2006). Extending global mapping to biodiversity is
within reach, such that in the near future anyone will be able to
see and explore our knowledge of the distribution of all biological
diversity. Scientists, managers, educators and the general public
will simply load biodiversity layers into a virtual globe like Google
Earth™ or World Wind™ and explore biodiversity change over time,
developing ever increasing resolution as more reports of biodiversity
presences and absences, growth and loss are reported.
The map layers are not limited to showing raw data, but could also
present compiled information, such as estimated species richness,
behavioral and morphological characteristics, or proliferation of
lineages containing unique phenotypes or genotypes. Analyses can
be constantly run in the background to update this ever changing
map by accessing existing and evolving species data resources
(e.g. the Global Biodiversity Information Facility; http://gbif.org)
or genetic resources (e.g. GenBank, http://www.ncbi.nlm.nih.gov;
Barcode of Life, http://barcodinglife.org). This system would allow
users to follow the biodiversity changes happening within the scope
of their own queries for taxa, genes, populations, ecosystems,
geographic regions and times of interest. The global biodiversity
map will also be designed as an active agent, alerting users when the
diversity of species or ecosystems changes in unusual ways (Collins
et al., 2006). This map is therefore not static, but continues to grow
as new data are made available from continuing exploration of the
globe. The goal of this review article is to discuss how we might
build systems that allow us to first improve and then analyze and
visualize heterogeneous global biodiversity data and their change
over space and time.
© The Author 2009. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org 421

R.Guralnick and A.Hill
Fig. 1. Diagram showing the flow of data and tools necessary to create the evolving global biodiversity map. Distributed biodiversity data sources, including
observational data, digital archival data such as photographs, sounds or published literature data and natural history collections data, are published to a
biodiversity data portal in a common format encouraging aggregation and exchange. New data are constantly being fed into the portals. Georeferenced
biodiversity data can then be presented on the global biodiversity map. External analysis services may accumulate raw biodiversity data from the portal, and
associated genetic, phenotypic and environmental data, in order to create new views and new biodiversity knowledge, itself representable back on the map.
Each of these layers is available to interested users, who may develop their own workflows in order to track regions and taxa of interest.
2 BUILDING BIODIVERSITY REPOSITORIES
Before one can begin to summarize information and knowledge
about biodiversity into maps, or perform any analyses, we must
first make all legacy and current data about when and where species
occur available and easy to use. A crucial, but well known challenge
for biodiversity informaticians is to efficiently curate and distribute
these data; an effort that will lay the foundations of a global
biodiversity infrastructure (Guralnick et al., 2007).
A fundamental unit of biodiversity data, the point on the
map, is the species occurrence record. Primary occurrence data—
observations or vouchers of the units of biodiversity at specific
places and times—may seem to be a deceptively simple type of
record. However, such records have their complexities, especially
with regards to taxonomy and location. The natural history
collections community has invested significant effort to create
widely adopted minimum information standards (e.g. DarwinCore)
and transmission protocols (e.g. DiGIR and TAPIR) to standardize
and share species occurrence records (Graham et al., 2004). We
expect that in the next decade, at least a billion species occurrences
records will be made available through portals like the GBIF data
site (http://data.gbif.org) which as of now contains nearly 7500
collections representing more than 150 million occurrence records
(Edwards, 2004; Lane, 2006). Many of those records will have
high quality and up-to-date taxonomic data, standardized dates
of collection and computer-readable geographic coordinates and
associated spatial uncertainty measurements. These records will be
unlocked not only from natural history collections from around
the world, but also past and current surveys, from old and new
photographs documenting organisms and extracted records from the
soon to be completely scanned biodiversity heritage literature (see
Fig. 1; Sarkar, 2007; http://www.biodiversityheritagelibrary.org).
Data quality issues in regards to taxonomy, location and date of
collection remain. These data quality issues are a major impediment
to the wide-scale use of such records by scientists and other decision-
makers (Guralnick et al., 2007; Robertson, 2008; Yesson et al.,
2007). A major challenge in biodiversity informatics is to develop
methods and tools that can help increase data quality, thus making
records more usable for biodiversity research and in conservation
decision-making (Chapman, 2005). It is unreasonable to expect data
providers to deal in a timely manner with data quality issues on their
own. Manually cleaning data and improving data quality can be a
slow process, especially when each provider must learn the current
standards and best practices employed in the field. As well, many of
the data vetting operations are repetitive in nature, so it is inefficient
422

Biodiversity informatics
for each data provider to independently perform tasks necessary to
improve data quality.
One of the key operations for increasing the quality of data
available in our repositories and for ultimately building the
biodiversity map is converting locality descriptions that have
been recorded in multiple different formats (e.g. text descriptions
such as 5 km west of Boulder, CO) into standardized, computer-
readable latitude–longitude-geographic uncertainty measurement
triplets (Guralnick, 2006). Uncertainty, in particular, is essential
because it provides a basis for deciding the scale at which an
occurrence record can be used. If the uncertainty of a georeference
is <500 m radius, this might be suitable for local or regional studies.
A record with a geographic uncertainty >20 km may only be usable
at continental or larger scales (Guo et al., 2008).
The first step towards efficiently georeferencing occurrence
records is to adopt a process that will always return the same
latitude–longitude-coordinate uncertainty triplet for the same textual
string. The BioGeomancer service (http://www.biogeomancer.org/;
Guralnick et al., 2006) has done just this and is particularly useful
in conjunction with other georeferencing tools (e.g. GeoLocate,
Metacarta and Yahoo’s GeoPlanet). Users with minimal knowledge
of georeferencing process can quickly generate standardized
georeferences of high quality. They do not have to start from
scratch developing idiosyncratic methodologies that do not reflect
community developed best practices. However, even with automated
or semi-automated workbenches, the task of georeferencing over 1
billion biodiversity records in a reasonable amount of time is an
immense and daunting undertaking.
A partial solution to this challenge of large-scale georeferencing
and potentially other data quality challenges will be to not
only provide automation for one step of the process (e.g.
conversion process of textual or other locality formats to geographic
coordinates) but to create pipelines where those tools are constantly
operating on the growing set of biodiversity occurrence records
as they become available. This pipeline can be developed to
automatically feed digital occurrence data from providers to the
Biogeomancer service for georeferencing. The results of the process
are stored for providers to review and fold back into their original
records. In coordination with GBIF and BioGeomancer, this pipeline
is already being developed (http://biodiversity.colorado.edu/bgb/).
In its finished state, tools such as really simple syndication (RSS)
for human users and web services (e.g. REST or SOAP) for
programmatic access will provide automatic notification of progress.
By utilizing such a system, both the georeferencing rate and number
of adequately processed records of the world’s biodiversity data will
increase exponentially, thus allowing much faster use of more data
in biodiversity research.
Similar systems may be used in the future for different sets
of problems. The growing set of knowledge being assembled
by the Catalogue of Life (http://www.catalogueoflife.org/search.
php) and others (e.g. World Register of Marine Species, http://
www.marinespecies.org/) may be plugged in to similar pipelines
to provide valid taxonomic names to records spanning centuries
made available from data providers spanning the globe. Systems
like iSpecies (http://ispecies.org/) may automate linking primary
data with much of the existing web-based metadata, whereas
projects like EOL Encyclopedia of Life (Wilson, 2003; http://www.
eol.org/) may become the source of more detailed information
regarding species. By linking these projects, the points on the global
biodiversity map become rich with information and supplemented by
existing analyses and knowledge. Figure 1 summarizes how digital
biodiversity resources, repositories and analysis tools are linked
together.
3 DETERMINING SPECIES DISTRIBUTIONS
Species occurrences are the raw data used to generate many
different summary views on biodiversity. One essential product
generated from species occurrences are species distribution maps.
These maps document where species are present and absent
across their full range and are often used as input themselves
for further biodiversity measurements. Documenting the current
geographic distributions of taxa at a global scale is a demanding
task (Whittaker et al., 2005) and ultimately, the confidence in
distributional knowledge is a direct function of the chosen grain
size (Hurlbert and Jetz, 2007). At a spatial resolution of a continent
or the whole world, the quality of our distribution data for many
taxa is quite good. At the resolution of 10 × 10 km, the accuracy
of distribution data remains poor for all but the most well-studied
regions and taxa (Soberón et al., 2007). Furthering knowledge about
species distributions to finer resolutions will help maximize global
biological knowledge, improve the scientific basis for conservation
support and allow us to more accurately model changes in species or
population distributions that may occur under environmental change
scenarios.
Generating predictions of distributions can be accomplished using
a mix of presence–absence occurrence data, expert opinion and
GIS-modeling approaches. Occurrence data provides the outline of
species distributions but in almost all cases, there is ‘observer bias’
in terms of sampling (Graham et al., 2004) and this is especially
true in regions such as the tropics where sampling has been limited.
Thus there is a significant role for refining ranges based on expert
opinion of biologists familiar with the species and environments
in question, at least in some cases (Ferrier et al., 2002; Pearce
et al., 2001). Until now, such opinions were either codified in the
literature or derived directly from the experts themselves. The rise of
social networks provides a third option for storage of expert opinion
distribution knowledge. Such knowledge can be compiled in social
networking applications like Scratchpads (http://scratchpads.eu) that
allow annotations of species ranges based on knowledge of the
species and environment.
The next step is to combine expert-based extent of occurrence
data with ecological niche modeling. Ecological niche modeling
is a rapidly emerging area that utilizes available presence only or
presence–absence data along with environmental data (e.g. climate
models and remote sensed data) to quickly output predictions about
suitability of habitat in areas that have not been sampled (Soberón
and Peterson, 2004) in GIS layer formats. As more data becomes
available to global repositories, and as environmental data layers
become finer scale, the more tractable finer scale ecological niche
modeling becomes. Although niche modeling is both quantitative
and repeatable, this approach does not yet take into account biotic
interactions and other biogeographic factors that may limit species
ranges (Guisan and Zimmerman, 2000; Soberón and Peterson,
2005). Thus it is the combination of modeling approaches and
expert opinion information that will put us into the position to make
accurate, scalable maps of species distributions (McPherson et al.,
2006; Pearce et al., 2001) that can be easily visualized on flat maps
423

R.Guralnick and A.Hill
or virtual globes. These modeled range maps represent a central
information product for our global biodiversity map.
Once the relationship between species occurrences and
environmental parameters is known in the present, it is possible
also to make predictions about future (e.g. Hijmans and Graham,
2006), and even past (e.g. Martínez-Meyer et al., 2004; Waltari
et al., 2007; Yesson and Culham, 2006) distributions if species niches
are conserved. Such forward or back-casted species distributional
modeling only requires one further set of environmental data—past
and future environmental landscapes (such as IPCC modeled climate
scenarios under different models of CO2 use; (http://www.ipcc-
data.org/ar4/gcm_data.html). Such models provide repeatable,
quantitative outputs concerning how species distributions may
change in location and size as climates or landscapes change. This
approach may be particularly warranted when it is known that
species distributions have shown consistent responses to climate
changes from the recent past to the present (Kharouba et al.,
in press). Because the outputs of models are also ‘maps’, these
predictions of distribution in the present and future are also part
and parcel of our larger global biodiversity map. There are already
persistent pipelines in place to rapidly handle many of the modeling
tasks as new and better data and modeling approaches become
available (LifeMapper, http://www.lifemapper.org; Stockwell et al.,
2006; ), although there is still a need for long-term archiving of
results. New pipelines could be designed by users to automate
new species distribution maps based on the best new data and
algorithms.
4 PIPELINES FOR BIODIVERSITY ASSESSMENT
The combination of finer scale range maps, data harvesting and
analysis pipelines can be easily modified to tackle new sets of
questions. The first will likely be simple but crucial questions such
as ‘How well sampled is biodiversity in a geographic region of
interest’ and ‘where should we go to sample more’. As of now, the
approach to answering these questions requires a huge investment
in data collection and individual analyses, which may or may not be
published and which are often quite difficult to repeat. Developing
persistent pipelines to answer these questions, and then housing
the results along with original data could radically change current
practices in biodiversity research. For example, in Guralnick et al.
(2007) we showed how simple data harvesters can be built to collect
data that meet certain specifications from data portals, which are then
stored in local databases, placed in grid cells representing different
geographic areas and automatically sent to analysis services for
further processing to estimate key biodiversity measures like species
richness. We then demonstrated how this can help us understand
biodiversity sampling quality in different regions of the world,
from southern Africa, to eastern North America, for groups like
birds and mammals. Further, these map outputs can become a
growing part of our global biodiversity map, representing summary
information aggregated from either individual species range maps
or occurrence data.
As these tools are set up to automatically harvest information
from data portals and perform analyses, they can be reassembled
and redeployed as new data and information flows into biodiversity
repositories (Fig. 1). Users themselves will be able to set up
feeds for taxa of interest in regions of interest, and receive
notifications when new records or analyses have become available.
As they do, users can redeploy their tools, or ones from a growing
library of tools made available as services, and determine if
new information (e.g. occurrence data, environmental data, niche
models) changes previous assessments of biodiversity or suggests
changes to biodiversity patterns that might be worth further study.
Therefore any and all analyses of regional, continental or even
global biodiversity can be run again and again to see how continued
aggregation of new data helps provide a well sampled view of
biodiversity of all organisms. Although this seems like a very
ambitious goal, it is not out of reach. The data, knowledge and
analysis approaches needed to document global biodiversity are
available, and harvesting tools and notification tools such as RSS
are simple enough to deploy.
5 INTEGRATION WITH PHYLOINFORMATICS
Although much information can be gleaned from just the
sampled distribution of organisms across the landscape (Funk
and Richardson, 2002; Kress et al., 1998), there are also limits.
Patterns of species distribution have come about through historical
processes, and inferring these processes through examination of trait
and molecular variation and sorting of this variation into distinct
lineages provides a much richer set of information about biodiversity
(Wheeler, 1995).
In biodiversity research, a synthesis of disciplines and approaches
drawing from multiple areas in ecology and phylogenetic
systematics is underway, as evidenced by integration of phylogenetic
approaches into community ecology and species distribution
modeling approaches into molecular ecology. More and more
frequently, diversity is measured not simply as number of
species, but instead utilizing a measurement based on phylogenetic
relatedness. Phylogenetic diversity (PD) is calculated as the sum
diversity of the overall phylogenetic branch lengths for a set of
genes or taxa in a sampled region (Faith, 1992). Such measurements
better reflect the diversity of unique traits that might exist within
an area. At the population level, back-casted species distribution
modeling approaches can be combined with phylogenetic and
phylogeographic methods as complementary datasets in order to
provide stronger test for past distribution and demography changes
that can be used to calibrate how changes may happen in the future
(Carstens and Richards, 2007; Waltari et al., 2007).
A current integrative approach to documenting population
histories is to construct potential past distributional models, as
discussed above, as a way to generate initial hypothesis that can
be formally compared with genetic data (Carstens and Richards,
2007). If there is concordance between results, then one can
also run projections of species or lineage distributions into the
future utilizing models of climate change. By extrapolating from
single species to multi-species studies in a region, it is possible
to examine concordance of patterns between species in a region
that might suggest action of concerted environmental forces acting
to shape community level genome variation. When combined with
the burgeoning growth in metagenomic, genomic and proteonomic
approaches, we can track the diversity and distribution of functional
genes and their products across landscapes. Finally, we can relate
variation in those patterns to known environmental changes, thus
generating a more process-oriented understanding of biodiversity
patterns.
424

Biodiversity informatics
Below we present two case studies that integrate biodiversity
informatics approaches with those from phyloinformatics and
other bioinformatics subdisciplines. These case studies show how
automation tasks can be employed to gather and process datasets
usable for biodiversity assessment. In the first case study, we
develop an automated pipeline for measuring species conservation
priority that utilizes both phylogenetic uniqueness and extinction
risk. The second case study uses a similar pipeline approach, but
focuses on analysis of full genomes of influenza A viruses in order
to document continuing evolution and geographic spread of this
disease.
6 PD CASE STUDY
Our first case study utilizes data made available from the Consortium
for the Barcode of Life (CBOL) to automate assessment of PD
in relation to conservation scenarios. CBOL is a global initiative
to utilize short DNA sequences from a standardized and agreed-
upon position in the genome as a molecular diagnostic for
species-level identification (Savolainen et al., 2005). CBOL has
developed a web-based Barcode of Life Data System (BOLD;
http://www.barcodinglife.org/views/login.php) that provides access
to all collected sequence data. There are currently ∼500 000 barcode
records available through BOLD, and 49 178 formally described
species using barcode data. Although BOLD is often used for species
identification, Faith and Baker (2006) note the utility of barcode
data for PD assessment as well. A web service will soon provide
programmatic access to barcode sequence data.
Here we utilize sequence data from the Barcode of Life project
‘Birds of North America’, in order to assess species conservation
priorities. This dataset was selected because it is particularly
well sampled, including multiple sequences for many of the
species in the region of interest (2586 sequences representing
657 bird species along with an outgroup crocodilian sequence
for rooting). Work by Kerr et al. (2007) documents the utility
of barcoding in documenting species level identification in North
American birds. As well, the coverage of species is excellent,
representing over 90% of the breeding and pelagic birds in Canada
and the United States. These barcodes predominately come from
vouchered museum specimens (Kerr et al., 2007). Finally, 23
species included in the BOLD dataset are listed as vunerable,
endangered or critically endangered based on IUCN red list
criteria (http://www.iucnredlist.org/static/categories_criteria_3_1).
The outcome of our analysis is a quantification of conservation
priority based on phylogenetic history and extinction risk (as
discussed in Faith, 2008) for all the sampled birds in North America.
The approach used here is simple: datasets are automatically
accumulated from the BOLD website. Unfortunately, a full toolkit
of external connectivity tools for BOLD is not yet fully operational
but synchronization services to share prealigned sequence data and
metadata required for this study are up and running. After also
collecting data about the specimens from which sequences were
drawn (e.g. taxonomy, geographic information, date of collection),
the next step is to perform tree-building.
We performed phylogenetic analyses using a RAxML
(Stamatakis, 2006) implementation available through a REST
API service from the Cyberinfrastructure for Phylogenetic Research
(CIPRES) project. In the future, it will be possible to include
other potential user-specified analyses (e.g. parsimony or Bayesian
Fig. 2. Conservation values for seven endangered (EN) or critically
endangered (CR) bird species in North America calculated based on
extinction threat and evolutionary distinctness as implemented by the
HEDGE and EDGE algorithms. Full results for all species, including
those species listed as vunerable on IUCN red list is available online at:
http://biodiversity.colorado.edu/bi/.
analysis). Trees are gathered and stored as a variable within the
PHP script.
To assess the conservation priority of the data, we ran three
analyses: RM (Redding and Mooers, 2006); EDGE (evolutionarily
distinct and globally endangered; Isaac et al., 2007); and HEDGE
(Steel et al., 2007). These three different analyses are different
ways to measure phylogenetic distinctness (see Faith, 2008 for more
details). We require an additional piece of information to run these
analyses: species extinction probability. To create these we used the
IUCN red list ratings, ranging from ‘vunerable’ to ‘endangered’ to
‘critically endangered’, representing risk of extinction. We assigned
critically endangered species a 99% chance of extinction in the
next 100 years, endangered species a 50% chance of extinction,
vulnerable species a 10% chance and least concern and near
threatened species a 1% chance, generally following values in
Mooers et al., 2008. Accumulating the red list information was
done manually using data available from Birdlife International
(http://www.birdlife.org/datazone/species/index.html), but this too
is a process that could be incorporated into the pipeline itself. Each
of these files is read and stored as arrays within a PHP script. Sample
files are included at http://biodiversity.colorado.edu/bi/.
We generated calculations for RM, EDGE and HEDGE for each
of the species from BOLD based on the phylogeny and extinction
risk. All results are stored within arrays for further manipulation
and output. At http://biodiversity.colorado.edu/bi/ we include a
sample output script that utilized the Google Charts API to
generate a graph of RM, EDGE and HEDGE results. The results
of the EDGE and HEDGE analysis for the seven endangered
or critically endangered species are shown in Figure 2. The
critically endangered species Brachyramphus brevirostris (Kittlitz’s
Murrelet) has the highest EDGE and HEDGE values. Among the
endangered species, Pterodroma hasitata (Black-capped Petrel),
Oceanodroma homochroa (Ashy storm-petrel) and Brachyramphus
marmoratus (Marbled Murrelet) rank next in both EDGE and
HEDGE measurements. These four species might therefore be
given priority for conservation over the other endangered species
Phoebastria nigripes, Grus americana and Centrocercus minimus.
Gymnorhinus cyanocephalus (Pinyon Jay, a monotypic taxon) is the
425

R.Guralnick and A.Hill
highest scoring vunerable species based on both EDGE and HEDGE
results (results for vunerable species not shown).
The case study shows the ease in which sequence data from
any higher level taxon of interest in a region of interest can be
accumulated from BOLD, analyzed and used along with extinction
risk to make assessment of species conservation priorities. Such
data could also be used for many other kinds of biodiversity
assessment, from determining regional endemism to measuring
overall community richness. This case study could be easily
extended so that much of the process is automated. Such
automation would provide updated knowledge of conservation risk
for all organisms as more and more data from BOLD becomes
available. This will provide means for unprecedented exploration
of evolutionary patterns in biodiversity and ultimately conservation
decision making.
7 TRACKING DISEASE EVOLUTION CASE STUDY
The second case study uses much of the same machinery as presented
above, only in this case we are tracking diversity of influenza A
viruses and how this diversity may have impacts on human health.
Unlike the above case study example, we are less interested in
overall PD in different regions of the world, although the same
approach could be applied to generate those values. Instead, we
want to understand how influenza A strains are proliferating across
the landscape. Although the focus is on influenza A viruses for this
case study, the system can be applied to other biological infectious
agents that are of concern.
In order to document the continuing evolution of viruses, we
have built an information system built around a set of operations
that we want to rapidly and automatically perform in order
to provide near real-time results for monitoring purposes. The
process begins with periodically harvesting genomic data coming
in from flu sequencing projects occurring around the globe (e.g.
http://www3.niaid.nih.gov/research/resources/mscs/Influenza/). In
the last 10 years, the number of sequenced influenza A flu genomes
has grown by orders of magnitude and that rate will likely continue
or accelerate into the future. From there, we automate construction of
alignments and phylogenetic analyses of these genomes, along with
automated georeferencing of the locations where the genomes were
isolated. Finally, we provide information on which types of animals
are carrying which lineages and how the virus might be transmitted
among different host types. This information is particularly crucial
for documenting increased transmissibility of strains like avian
influenza H5N1, now a predominately avian disease, from birds to
mammals.
Our workflow is designed to accumulate, link and reformat
sources relevant to disease evolution and ecology. Currently
we utilize only two main data sources: NCBI’s GenBank for
genomic data and metadata about viral isolation, and external
gazetteers (e.g. the BioGeomancer workbench) and the Getty
Thesaurus of Geographic Names (http://www.getty.edu/research/
conducting_research/vocabularies/tgn/index.html) for place names.
From these genetic and geographic sources, we can gain sufficient
data to reconstruct both relationships among the viral isolates and
their location on the globe, and then package that information for the
user. Figure 3 summarizes the steps in creating meaningful outputs
that can be used for disease monitoring (see Hill and Guralnick,
2008 for full workflow diagram). It is important to note that there are
Fig. 3. Influenza monitoring workflow: (1) data harvested from Genbank; (2)
data aligned, trimmed and a phylogeny run through the Cipres web service;
(3) geographic coordinates and phylogeny used to construct a geophylogeny
in KML for viewing in GoogleEarthTM; (4) outputs made available for users
to download and reuse.
numerous places to utilize a distributed network of environmental
and biological data resources and methods of analyses. We touch on
a few of the obvious distributed resource and analysis tools that can
be linked to our application below.
Because our system is currently designed for influenza A,
data harvesting is generally straightforward. Influenza genetic and
associated data already organized and stored at the NCBI ftp site.
Therefore, data harvesting is a simple programmatic process of
running cURL to transfer the files to a local repository. Following
the transfer, the files are parsed into a local PostgreSql database. The
parsed data includes the full genome sequence, date of collection,
location of collection, the host from which the genome was isolated
and any other metadata that might be of interest. In order to allow for
mapping, we then tie each isolate to the decimal degree latitude and
longitude for its best locality description. In order to link locality
strings to latitude and longitude, we currently use a hand assembled
library of common localities of influenza isolation.
The phylogenetic analysis steps in the process are similar to those
described for first case study but first the genomic data needed to
aligned. We align each dataset using MUSCLE (http://www.drive5.
com/muscle/) and MAFFT (http://align.bmr.kyushu-u.ac.jp/mafft/
software/) services available through the European Bioinformatics
Institute (EBI; http://www.ebi.ac.uk/). For the time being, we are
currently only running analyses on influenza sampled with full
genomes sequenced (those represented by all eight vRNAsegments).
Alignments are programmatically trimmed, simply removing as
many leading and trailing columns as necessary until a certain
threshold of the data are represented in the column. Trimmed
alignments are stored in the PostgreSQL database. After alignment,
we used the RAxML implementation available through a REST API
service from the CIPRES project to generate maximum likelihood
trees.
After phylogenies are returned from CIPRES, we combine
phylogeny and geographic information to form the structure of
the Keyhole Markup Language (KML) visualization of disease
spread viewable in GoogleEarthTM. In short summary, each terminal
taxon and hypothetical taxonomic unit (HTU) was assigned a set
of geographical coordinates and altitude. Altitude is based on the
phylogenetic depth of the node, where terminal taxa rest on the
surface of the globe and HTUs are shown floating above the globe—
the farther towards the base of the tree, the higher the altitude above
426

Biodiversity informatics
Fig. 4. Visualization showing a phylogenetic tree of H5N1 avian influenza
as projected onto GoogleEarthTM. The visualization is one of the outputs
made available through the workflow in Figure 3.
the globe the HTU. For more details on creating these KMLs,
see Janies et al. (2007). By utilizing metadata on isolation date
and the KML timespan function, we can add an explicit temporal
component to the GoogleEarthTM visualization allowing users to
track lineage movement over time. Figure 4 presents a snapshot
visualization for H5N1 influenza strains. Outputs from our workflow
are available here: http://biodiversity.colorado.edu/ipl/. From the
website, users can download original datasets, making it possible
for them to use our preliminary analyses to formulate hypotheses
and predictions, and then refine the datasets to test those predictions.
Also available are files from each stage of the analysis, including
alignment, and RAxML trees, and KML files.
We foresee many future directions to expand our data visualization
and analysis workflow. This includes using analysis of codons to
provide information about functional changes to the virus. Such
a process would be done by programmatically reconstructing the
coding sequence from nucleotide sequences, and analyzing those
through other programs (i.e. HyPhy’s http://datamonkey.org). An
area of particular interest for monitoring would be the automation
of the work done by Hill et al. (in press) detecting codon-level
selection for known mutations that change how the virus functions
(e.g. resistance to antiviral drugs).
This case study has a different emphasis in terms of visualization
and analysis compared with our first case study. Here, we are less
concerned about providing an assessment of conservation priority
for a set of taxa in a region of interest, and more focused on continued
monitoring, analysis and visualization of how strains of diseases and
the mutations they are carrying spread across the landscape. This
case study also examines biodiversity—of influenza viruses—and
could grow to include how these diseases are both transmitted by,
and affect, other vertebrate groups.
8 CHALLENGES, SOLUTIONS AND
CONCLUSIONS
We will face numerous challenges along the path to unifying
biodiversity data, information and knowledge. One challenge we can
address quickly is the lack of information regarding the structure
of the growing biodiversity informatics network. Where are all
the sources of digital primary occurrence records and how are
they being provided? What online services have already been
designed to analyze these data and what mechanisms are in place
for programmatic access? What are all the sources of freely
available environmental data? What are the terms of data use and
is appropriate metadata for attribution of data available? What
pipelines are already in place and how are they storing their results?
These questions are currently difficult to answer. Much as physical
collections have catalogs that map their many specimens, the digital
collections, services and pipelines network needs a similar map
or registry. At each node of the network, it is necessary that we
know the format and method by which data are provided, how
it is queried, or how a service is run. This will allow new users
and interested groups to immediately locate the data they need, the
service they want to run, and determine how they should design
their pipeline to make it work within the network. The map should
also allow new users to document their own emerging nodes and
pipelines. There will be many benefits to mapping this system
while it is still young such as: the reduction in duplicated services;
improved time to develop new services; and rapid identification
of areas where new pipelines or services are still needed. The
Biodiversity Information Standards ( http://www.tdwg.org/) group
has begun to assemble many of the network nodes, through both their
‘Biodiversity Information Projects of the World’ and ‘Biodiversity
Information Networks Database’, yet only fraction of existing points
have been documented while such services exponentially increase
in number.
Another challenge that hinders the union of biodiversity
informatics with other informatics fields is that different data from
a single organism may be stored in several different collections.
When individual records are also associated with global identifiers
(GUIDs), it becomes possible to automatically harvest such records
from multiple access points (Page, 2008). Such automation provides
the means for faster data acquisition and analysis in order document
changing genetic, species and ecosystem diversity.
The global, unified and continually evolving biodiversity map
will be multilayered and multidimensional, representing raw data,
summary data and information and knowledge. Using this map
will ultimately allow us to make better decisions about how to
understand and manage biodiversity in a rapidly changing world.
We argue it is already possible and will soon become commonplace
for researchers to acquire knowledge, in near-real time, about not
only just the changing patterns of biodiversity, but also processes
that might be driving those changes. A process oriented view is
going to require integration of multiple data types and sources. As
the pipelines pushing data that allow us to perform these analyses
continue to widen, informatics approaches that are integrative and
take from the best of traditional bioinformatics, phyloinformatics
and biodiversity informatics will be needed to achieve the goal of
such a global biodiversity map.
ACKNOWLEDGEMENTS
We appreciate the help of David Schindel and Sujeevan
Ratnasingham from the Consortium for the Barcode of Life and
Barcode of Life Data Systems for their help in developing our case-
study example automating phylogenetic diversity measurements.
427

R.Guralnick and A.Hill
Our rationale for developing scalable species distribution services
developed through discussions with Walter Jetz.
Funding: Global Biodiversity Information Facility (#2007-92 to
R.P.G. and A.H.); National Biological Information Infrastructure
(04HQAG0121 to R.P.G.); National Science Foundation (0110133
to R.P.G.); Gordon and Betty Moore Foundation.
Conflict of Interest: none declared.
REFERENCES
Bisby,F.A. (2000) The quiet revolution: biodiversity informatics on the internet. Science,
289, 2309–2312.
Butler,D. (2006) Mashups mix data into global service. Nature, 439, 6–7.
Canhos,V.P. et al. (2004) Global biodiversity informatics: setting the scene for a “new
world” of ecological forecasting. Biodiversity Informatics, 1, 1–13.
Carstens,B.C. and Richards,C.L. (2007) Integrating coalescent and ecological niche
modeling in comparative phylogeography. Evolution, 61, 1439–1454.
Chapman,A.D. (2005) Principles and methods of data cleaning. Report for the Global
Biodiversity Information Facility 2004. GBIF, Copenhagen, Denmark.
Collins,S.A. et al. (2006) New opportunities in ecological sensing using wireless sensor
networks. Front. Ecol. Environ., 4, 402–407.
Edwards,J.L. (2004) Research and societal benefits of the global biodiversity
information facility. BioScience, 54, 485–486.
Faith,D.P. (1992) Conservation evaluation and phylogenetic diversity. Biol. Conserv.,
61, 1–10.
Faith,D.P. (2008) Threatened species and the preservation of phylogenetic diversity
(PD): assessments based on extinction probabilities and risk analysis. Conserv.
Biol., 22, 1461–1470.
Faith,D.P. and Baker,A.M. (2006) Phylogenetic diversity (PD) and biodiversity
conservation: some bioinformatics challenges. Evol. Bioinform. Online, 2, 70–77.
Ferrier,S et al. (2002) Extended statistical approaches to modelling spatial pattern
in biodiversity in north-east New South Wales. I. Species-level modelling.
Biodiv. Conserv., 11, 2275–2307.
Funk,V.A. and Richardson,K.S. (2002) Systematic data in biodiversity studies: use it
or lose it. Syst. Biol., 51, 303–316.
Graham,C.H. et al. (2004). New developments in museum-based informatics and
applications in biodiversity analysis. Trends Ecol. Evol. (Amst.), 19, 497–503.
Guo,Q. et al. (2008) Georeferencing locality descriptions and computing associated
uncertainty using a probabilistic approach. Int. J. Geogr. Inf. Sci., 22, 1067–1090.
Guisan,A. and Zimmermann,N.E. (2000) Predictive habitat distribution models in
ecology. Ecol. Modell., 135, 147–186.
Guralnick,R.P. et al. (2006) Biogeomancer: automated georeferencing to map the
world’s biodiversity data. PLoS Biol., 4, 1908–1909.
Guralnick,R.P. et al. (2007) Toward a collaborative, global infrastructure for biodiversity
assessment. Ecol. Lett., 10, 663–672.
Heywood,V.H. and Watson,R.T. (1995) Global Biodiversity Assessment. Cambridge
University Press, Cambridge. pp. 1–1152.
Hijmans,R.J. and Graham,C.H. (2006) The ability of climate envelope models to predict
the effect of climate change on species distributions. Glob. Chang. Biol., 12, 1–10.
Hill,A.W. and Guralnick,R.P. (2008) A case study for distributed systems and automated
biodiversity informatics: genomic analysis and geographic visualization of disease
evolution. In Gray,A. et al. (eds) British National Conference on Databases
Springer-Verlag Lecture Notes in Computer Science series 5071. Springer, Berlin
and Heidelberg, pp. 270–279.
Hill,A.W. et al. Evolution of drug resistance in multiple distinct lineages of H5N1 avian
influenza. Infect. Genet. Evol. (In press).
Hurlbert,A.H. and Jetz,W. (2007) Species richness, hotspots, and the scale dependence
of range maps in ecology and conservation. Proc. Natl Assoc. Sci., 104,
13384–13389.
Isaac,N.J.B. et al. (2007) Mammals on the EDGE: Conservation for sustainable
development. IUCN/United Nations Environment Program/World Wildlife Fund,
Gland Switzerland.
Janies,D. et al. (2007) Genomic analysis and geographic visualization of the spread of
avian influenza. Syst. Biol., 56, 321–329.
Jenkins,M. (2003) Prospects for biodiversity. Science, 302, 1175–1177.
Kerr,K.C.R. et al. (2007) Comprehensive DNA barcode coverage of North American
birds. Mol. Ecol. Notes, 7, 535–543.
Kharouba,H.M. et al. Historically calibrated predictions of butterfly species’ range shift
using global change as a pseudo-experiment. Ecology (In press).
Kress,W.J. et al. (1998) Amazonian biodiversity: assessing conservation priorities with
taxonomic data. Biodiv. Conserv., 7, 1577–1587.
Lane,M. (2006) Information infrastructure for global biological networks. Microbiol.
Aust., 27, 23–25.
Loreau,M. et al. (2006) Diversity without representation. Nature, 442, 245–246.
Martínez-Meyer,E. et al. (2004) Ecological niches as stable distributional constraints on
mammal species, with implications for Pleistocene extinctions and climate change
projections for biodiversity. Glob. Ecol. Biogeogr., 13, 305–314.
McPherson,J.M. et al. (2006) Using coarse-grained occurrence data to predict species
distributions at finer spatial resolutions—possibilities and limitations. Ecol. Modell.,
192, 499–522.
Mooers,A.Ø. et al. (2008) Converting endangered species categories to probabilities of
extinction for phylogenetic conservation prioritization. PLoS ONE, 3, e3700.
National Academy of Sciences (2001) Grand Challenges in Environmental Sciences.
National Academy Press, Washington DC, pp 1–106.
Novacek,M. and Cleland,E.E. (2001) The current biodiversity extinction event:
scenarios for mitigation and recovery. Proc. Natl Acad. Sci. USA, 98, 5466–5470.
Page,R.D.M. (2008) Biodiversity informatics: the challenge of linking data and the role
of shared identifiers. Brief. Bioinformatics, 9, 345–354.
Pearce,J.L. et al. (2001) Incorporating expert opinion and fine-scale vegetation mapping
into statistical models of faunal distribution. J. Appl. Ecol., 38, 412–424.
Pimm,S.L. et al. (1995) The future of biodiversity. Science, 269, 347–350.
Redding,D.W. and Mooers,A.O. (2006) Incorporating evolutionary measures into
conservation prioritization. Conserv. Biol., 20, 1670–1678.
Robertson,D.R. (2008) Global biogeographical data bases on marine fishes: caveat
emptor. Divers. Distrib., 14, 891–892.
Sarkar,I.N. (2007) Biodiversity informatics: organizing and linking information across
the spectrum of life. Brief. Bioinformatics, 8, 347–357.
Savolainen,V. et al. (2005) Toward writing the encyclopaedia of life: an introduction to
DNA barcoding. Philos. Trans. R. Soc. Lond., B, Biol. Sci., 360, 1805–1811.
Soberón,J. and Peterson,A.T. (2004) Biodiversity informatics: managing and applying
primary biodiversity data. Trans. R. Soc. Lond., B, Biol. Sci., 359, 689–698.
Soberón,J. and Peterson,A.T. (2005) Interpretation of models of fundamental ecological
niches and species’ distributional areas. Biodiv. Inform., 2, 1–10.
Soberón,J. et al. (2007) Assessing completeness of biodiversity databases at different
spatial scales. Ecography, 30, 132–150.
Stamatakis,A. (2006) RAxML-VI-HPC: maximum likelihood-based phylogenetic
analyses with thousands of taxa and mixed models. Bioinformatics, 22, 2688–2690.
Steel,M. et al. (2007) Hedging our bets: the expected contribution of species to future
phylogenetic diversity. Evol. Bioinform., 3, 237–244.
Stockwell,D.R.B. et al. (2006) The use of the GARP genetic algorithm and Internet grid
computing in the Lifemapper world atlas of species biodiversity. Ecol. Modell., 195,
139–145.
Waltari,E. et al. (2007) Locating pleistocene refugia: comparing phylogeographic and
ecological niche model predictions. PLoS ONE, 2, e563.
Wheeler,Q.D. (1995) Systematics, the scientific basis for inventories of biodiversity.
Biodiv. Conserv., 4, 476–489.
Whittaker,R.J. et al. (2005) Conservation biogeography: assessment and prospect.
Div. Distrib., 11, 3–23.
Wilson,E.O. (1988) The current state of biological diversity. In Wilson,E.O. (ed.)
Biodiversity. National Academy Press, Washington, DC, pp. 3–18.
Wilson,E.O. (2000) A global biodiversity map. Science, 289, 2279.
Wilson,E.O. (2003) The encyclopedia of life. Trends Ecol. Evol. (Amst.), 18, 77–80.
Yesson,C. and Culham,A. (2006) Phyloclimatic modeling: combining phylogenetics
and bioclimatic modeling. Syst. Biol., 55, 785–802.
Yesson,C. et al. (2007) How global is the global biodiversity information facility? PLoS
ONE, 2, e111.
428