Unifying Phenotypes to Support Semantic Descriptions

Abstract

In life sciences, there are several biological datasets shared through the web. All this abundance of data carries a great opportunity to explore complex relationships among the diversity of species. However, their physical format varies from independent data files to databases, which are heterogeneous in model and representation, hampering their integration. Ontologies are one of the promising choices to address this challenge. However, the existing digital phenotypic descriptions are stored in semi-structured formats, making extensive use of natural language. If on one hand, this patrimony is highly relevant, on the other hand, converting it in ontologies is not a straightforward task. The present article addresses this problem adding an intermediate step between semi-structured phenotypic descriptions and ontologies. It remodels semi-structured descriptions to a graph abstraction in which the data are linked. Graph transformations subsidize the transition from semi-structured data representation to a more formalized representation through ontologies.

1. Introduction

Bioinformatics is the science of integrating, managing, mining and interpreting information from biological data [ Gibas and Jambeck 2001 ]. In the life science field, there are a large number of distributed biological datasets freely available and ready to use. However, this wealth of information has hardly been tapped even today due its distributed nature, heterogeneity and complex data types and representation [ Parr et al. 2012 ]. In this scenario, their combination and interconnection are barely feasible [ Quan 2007 ]. A massive amount of relevant information is hidden in the potential connection of unrelated files.

In this work we are interested in a specific biology context, in which biologists apply computational tools to build and share digital descriptions of living beings as phenotypes. These descriptions are a fundamental starting point for several biology tasks, like living beings identification and tools for phylogenetic tree analysis. Even though the last generation of these tools is based on open standards (e.g., XML), the descriptions are still based on textual sentences in natural language [ Balhoff et al. 2010 ].

Semantic integration in this context is one of the main challenges. Besides ontologies to support phenotype description, there are tools to annotate descriptions by associating ontology concepts to textual descriptions [ Balhoff et al. 2010 ]. This distinction between description and their annotations based on ontologies does not consider that descriptions can conversely contribute to ontology expansion and revision. The challenge in this work is to establish a model to represent a common denominator among phenotipical description standards, which will support findings in the latent semantics implicit in relations in a strategy inspired by folksonomies. These semantics can guide the interaction between textual descriptions and ontologies.

In a previous work [ Alves and Santanche` 2013 ], we showed that the latent semantics presented in tags and their correlations, as a product of an organic work collectively produced by a community on the web (the folksonomies), can be exploited to expand and review ontologies. While the model behind folksonomies is based on the correlation of three elements – tags, resources and users – descriptions in the biological context present a more complex and specialized structures. Co-occurrence is a strong principle we considered to extract latent semantics. The main idea is that the set of tags put together in a given resource can provide a “context” to interpret each tag. Consider a tag cell, which can have a distinct interpretation according to the context. The co-occurrence with the tags cytoplasm or organelle will put it in the biology context. Moreover, the compilation of data concerning the occurrence and co-occurrence of millions of tags can support the analysis of similarity among terms – see more details in [ Alves and Santanche` 2013 ]. We consider that we can apply an equivalent technique to put terms of phenotype descriptions in a context, to improve their interpretation and correlation.

The present paper addresses this problem in exploiting existing biology assets related to phenotypic descriptions, and the latent semantics resulting from their interconnection, to support their development towards a richer semantical representation, as part of ontologies. It implies promoting relations among concepts to first class citizens. Accordingly, we designed a three layered method illustrated in Figure 1, in which graph databases intermediate this evolvement process from fragmentary data sources to accomplish full integration descriptions as ontologies.

Our approach remodels semi-structured descriptions to a graph abstraction, in which the data can be integrated more easily. Graph transformations are applied for the transition from a semi-structured data representation to a more formalized representation through ontologies. As we will further explain, this graph representation will also support an analytical tool to compare data across studies, wherein it will help evolutionary biologists to answer evolutionary questions. This paper presents a work in progress concerning the first step of this method, focusing in the integration of data from the semistructured data layer and their transition to the graph data abstraction layer. Our proposed graph-based model is derived from a comparative analysis among four standards related to phenotype description, plus a practical experiment.

Ontology concepts

Graph data abstraction

Semi-structured data

This paper is organized as follows: Section 2 summarizes the related work; Section 3 presents the comparative analysis which subsidizes our minimal common denominator model; Section 4 presents out graph-based model; Section 5 shows a practical experiment of unifying phenotypes; Section 6 presents concluding remarks.

2. Related Work

Integration is a key point as humans are progressively unable of handling the sheer volume of data presented [ Bell et al. 2009 ]. It is an important step towards knowledge discovery [ Lenzerini 2002 ]. The integration of digital phenotype descriptions is a relevant challenge in this context since they support fundamental biology tasks as the building of identification keys for living beings and can support the creation of a complete evolutionary Tree of Life [ Parr et al. 2012 ] assembling genomic and morphological data so as to congregate the phylogenetic relationships among all living or extinct organisms [ Ciccarelli et al. 2006 ]. Likewise, integrating these data may contribute to better understanding of how a morphological trait became organized and evolved over time [ Mabee 2006 ].

Recent approaches enrich descriptions via ontology annotations, using the Entity-Quality (EQ) formalism for phenotype modeling. EQ is a representation [ Balhoff et al. 2010 ] which associates ontology entity terms (E) – e.g., bone or vertebra from Teleost Anatomy Ontology (TAO) – with quality terms (Q) – e.g., triangular, horizontal, smooth from the Phenotype and Trait Ontology (PATO) [ Dahdul et al. 2010 ]. Ontologies have gained wide acceptance in biology due to their ability of representing knowledge and also the advantage of querying and reasoning information [ Gkoutos et al. 2004 ]. Furthermore, semantic web standards to represent ontology concepts with unique identifiers facilitates interoperability across databases [ Mabee et al. 2007 ]. Recently, several tools have emerged to support annotation of biological phenotypes using ontologies, e.g., Phenex (http://phenoscape.org/wiki/Phenex) and Phenote (http://www.phenote.org/ ), both curation tools designed for annotation of phenotypic characters with ontology concepts using EQ formalism [ Balhoff et al. 2010 ].

[ Dahdul et al. 2010 ] developed a workflow for curation of phenotypic characters extracted from scientific publications. It is important to note the limitations of this curation process, considering that it is very time-consuming since it is manually carried out by domain experts.

3. Common Denominator

There is a wide variety of representation formats for phenotype description, adopted by information systems and open standards, which represent differently the same information. In this section, we analyze four of them – Xper2, SDD, Nexus and NeXML – looking for a minimal common denominator, which is the foundation for our graph-based model, to be used to link related information.

SDD, Nexus and NeXML are widely adopted open standards further detailed. Xper2 (http://lis-upmc.snv.jussieu.fr/lis/ ) is a management system adopted by the systematist community, for the storing, editing and analyzing of phenotype descriptive data. It focuses mainly on taxonomic descriptions, allowing creation, sharing and comparison of identification keys [ Ung et al. 2010 a, Ung et al. 2010 b]. Xper2 was developed in the Laboratoire Informatique & Syste´matique of the University Pierre et Marie Curie and this work is part of a bigger project in collaboration with this lab. Therefore, Xper2 was adopted for our practical experiments.

In order to illustrate our analysis, let us consider a practical case, in which a biologist is building a phenotype description of monitor lizards (genus Varanus). The process starts with the biologist collecting observations of lizards, organized as characters and character states (C, CS). [ Pimentcl and Riggins 1987 ] defined character as “a feature of organisms that can be evaluated as a variable with two or more mutually exclusive and ordered states”. The observations involved the species Varanus albiguralis and Varanus brevicauda. The final result is the character-by-taxon matrix illustrated in Figure 2.

Varanus albiguralis Varanus brevicauda 2 1 m fr o ' s l it r s o n l it a e h lsa fto r e n svn ito tra sce 1 2 2 1 s e l a c s l a h c u n

Nostrils' form

1 – well round 2 – oval or split-like

Transversal section of the tail

1 – laterally compressed 2 – roundish

Nuchal scales

1 – same size than head scales 2 – bigger than head scales

In order to transform these observations to digital records and generalize them – e.g., devising general characters and states observed in a genre of monitor lizards – the biologist will use a tool like Xper2. Phenotypes descriptions can be stored in the Xper2 native format or can be exported to the SDD open format. The Structure Descriptive Data (SDD) (http://wiki.tdwg.org/SDD) is a platform and application-independent XML-based standard developed by the Biodiversity Information Standards (historic acronym: TDWG) for recording and exchanging descriptions of biological and biodiversity data of any type [ Hagedorn 2007 ]. SDD is adopted by several other phenotype description tools – e.g., Lucid Central (http://www.lucidcentral.org) and Linnaeus II (http://www.eti.uva.nl/ ).

We further introduce some key elements of the SDD format, which are recurrent in the formats confronted in this section. A SDD description comprises, in a single file, a domain schema and its instances. Figure 3 shows a diagram with a fragment of a SDD file containing the description of a varanus lizard. A (C,CS) description in SDD has two main blocks: (i) defines the characters involved and their possible states – Figure 3 top; (ii) describes an Operational Taxonomic Unit (OTU) using the characters defined in (i) – Figure 3 bottom. OTU is a biology term which refers to a given entity in sampling level adopted to the study – e.g., a specimen, a gender etc.

<CategoricalCharacter>s and their <States> (shown in Figure 3 top) are primitives to describe an OTU [ Hagedorn 2007 ]. Each <CategoricalCharacter> has its <Representation> – comprising a label and a description as plain texts – and a set of <StateDefinition> elements with their possible states. <CategoricalCharacter> and <StateDefinition> elements defined here will be referred throughout the XML document by their ids.

The <CodedDescription> (Figure 3 bottom) links the OTU being described to States of each <CategoricalCharacter>. It has two essential items: (i) the OTU Datasets Dataset

id=“c6” CategoricalCharacter

id=“D1” CodedDescription

States SummaryData Representation

ref=“c6” Categorical

Label

“V. albiguralis” Detail “Monitors' nostrils may have different forms...”

id=“s12” StateDefinition

id=“s13” StateDefinition

Label

“well round” Detail

“Nostrils look like a quite per...” Label

“oval or split-like” Detail

“Nostrils are not perfectly rou...” State

ref=“s13” Detail

“White-throated monitor. Distribution: Africa (West...” being described, where its name and description are listed in natural language under <Representation>; (ii) a set of character and values (<Categorical> and <State>), which address the characters defined in the previous section through the ref attribute. It is possible and usual to define multiple states for a character of a given OTU. A first integration, problem observed here is that each character or OTU described does not have a global unique identification among documents. Therefore, the description can only be used by the document where it was declared and it is not possible to guarantee the equivalence of two or more <CategoricalCharacters>.

In Figure 5 we expand our analysis to the Xper2 native format, Nexus and NeXML. Our study addresses mainly morphological character descriptions. Figure 5 provides simplified diagrams focusing on the elements to record descriptions, which will be confronted here. Figure 4 presents the symbols adopted in the diagram. All the formats adopt XML and the symbols represent the relations among elements and their respective cardinality. Five types of elements, which are focus of our analysis, receive special symbols: the Entity being described, which can be a taxon or a specimen; the Character definition and its respective association with entities (Character instance); the State definition and its respective association with entities (State instance).

Nexus [ Maddison et al. 1997 ] is an extensively used file format developed for storage and exchange of phylogenetic data, including morphological and molecular characters, taxa distances, genetic codes, phylogenetic trees etc. It was designed in 1987 and it is still used by many popular software as Xper2 (http://lis-upmc.snv.jussieu.fr/lis/ ), Mesquite (http://mesquiteproject.org/ ), MrBayes (http://mrbayes.sourceforge.net/ ) and data repositories, like TreeBASE(http://treebase.org/ ) and Dryad (http://datadryad.org/ ). Nexus gathers together (C,CS) based descriptions and related trees [ Vos et al. 2012 ].

1Knowledge base of the genus Varanus from http://lis-upmc.snv.jussieu.fr/xper2/infosXper2Bases/listebases-recherche.php Element types structural element exclusive option

Relationship types one to one one to many (zero or more) Structural element specializations one to one (zero or one) one to many (one or more) Entity

NeXML (http://www.nexml.org) [ Vos et al. 2012 ] is a standard inspired by the Nexus. It supports and extends Nexus functionalities and addresses some Nexus limitations – e.g., connects objects with ontology concepts, supports citations and annotations [ Vos et al. 2012 ]. In order to accomplish full compatibility and interoperability among different environments, NeXML defines a formalized XSD grammar and enables semantic annotations of any element in a NeXML document, which goes towards to a “Minimum Information About a Phylogenetic Analysis” (MIAPA) standard.

These comparative diagrams show that even if the structures are arranged differently, they address the same key elements. All formats organize data in accordance with the (C,CS) data model that, in practice, is an entity-attribute-value (EAV) model, in which entities are OTUs, attributes are characters and values are character-states [ Vos et al. 2012 ]. Nexus and NeXML formats define a matrix, in which OTUs are listed in rows, characters are columns and the cells contain a numeric code for a specific characterstate (see Figure 2). Although Xper2 and SDD do not define a matrix, both formats have a similar structure to describe OTUs with their (C, CS) records.

4. From XML Structures to Graphs

The next step in our Three Tier Method is designing a graph model. In a previous work [ Alves and Santanche` 2013 ], we have compared several approaches to capture latent relations+semantics among tags produced collaboratively. Graph models to represent and analyze data were a common denominator. The role of the graph is not to reflect all details of the original model. The central challenge is how to abstract key elements, for which we are looking for potential relations to be discovered. It is a movement from the latent semantics to an explicit semantics expressed as links.

On one hand, we devised in the previous section the common denominator we are looking for: OTUs, character and character states. On the other hand, a second important ingredient is devising what is our target in ontologies. As mentioned in Section 2, a predominant ontology model for phenotype descriptions is the Entity-Quality (EQ) [ Balhoff et al. 2010 ]. An Entity refers to the “part” of the OTU being described, which is related to one or more Qualities. In a comparison with the (C, CS) approach, a Character comprises an Entity plus the Quality involved in the description in a single textual sentence. A State is a complementary part of the Quality. Even though it is not a trivial task to split Characters into their components of Entity and Quality, a first step will SDD

Xper2 Characters charStateLabel matrix

row

CategoricalCharacter CodedDescription taxon-­‐name

entry (a) Nexus

Representation

States Representation SummaryData (c) SDD

OTUs Format matrix Variables

Individuals charLabels stateLabels be linking disperse elements referring to the same semantic concept.

Departing from the key elements identified in the previous section, we can devise the following linking discovery challenges:

Which OTUs in the graph refer to the same real world OTU (link OTU-OTU)? Which characters can be applied to each OTU (link OTU-character)? Which states for each character can be observed in each OTU (link OTUcharacter-state)? Conversely, which OTUs have a given character+state? The answer to these questions will enable to integrate, summarize and compare data concerning each OTU and each character. Therefore, it becomes possible to answer queries like: What are the possible colors of a Varanus tongue? Which animals present an oval nostrils form?

The discovery process is carried by graph transformations. As graphs are crucial for our modeling approach, our method was built over graph databases. These databases reduce the gap between how data is modeled (as graphs) and how it is stored. It is capable of representing data structures with high abidance. Compared with relational databases, graph databases do not require join operations because it is done implicitly traversing the graph from node to node. Graph databases are less schema-dependent and for this reason, they can scale more easily in size and complexity as the application evolves.

The questions stated before were the basis to conceive the model presented in Figure 6. We adopted the property graph model, in which nodes and relationships can maintain extra metadata as a set of key/value pairs. Moreover, relationships are typed, enabling to create multi-relational networks with heterogeneous sets of edges. Different from single-relational networks, in which edges are of the same type, multi-relational networks are more appropriate to represent complex domain models, due the variety of relationship types in the same graph [ Rodriguez and Shinavier 2010 ].

In our graph model, OTUs and character-states are nodes connected by characters (edges). Therefore the statement “V. albiguralis has a well round tail shape” becomes V. albiguralis (node) ! tail shape (edge) ! well round (node).

Type Label Detail

OTU

OTU

Character Type Detail

Character

Character-State

State Type Label Detail

5. Practical Experiment of Unifying Phenotypes

We have implemented an automatic process to ingest SDD files into a graph database, in order to show the linking possibilities raised by our model. In our experiments, we use the Neo4j (http://www.neo4j.org/ ), an open-source graph database. Our data integration processing flow is divided into the main stages: preprocessing, data ingestion, data linkage.

One of the problems faced in bioinformatics is related to the identification of objects within and across repositories [ Page 2008 ]. More precisely, an object may refer to a taxon, gene, anatomical feature, phenotypic description, geographical location etc. Uniquely identifying those objects is undoubtedly a key point for the success of our proposed solution.

In order to address this issue, some organizations – e.g., Universal Biological Indexer and Organizer (uBio), Integrated Taxonomic Information System (ITIS), Catalogue of Life (CoL), The International Plant Names Index (IPNI), National Center for Biotechnology Information (NCBI) etc. – incorporated into their projects the Life Science Identifiers (LSIDs), which was proposed by the Object Management Group (OMG) (http://www.omg.org/ ). LSID is a persistent, location-independent resource identifier, whose purpose is to uniquely identify biological resources [ Clark et al. 2004 ]. The persistent property refers to the fact that LSID identifiers are unique, can be assigned to only one object forever and they never expire. The location-independent property specifies that each authority locally creates LSIDs and they are the responsible to guaranteeing the uniqueness of LSIDs.

We applied LSIDs to unify OTUs in the graph referring to the same real world object. In order to find a valid LSID, we adopted the Global Names Resolver (GNR) web service (http://resolver.globalnames.org/ ) that executes exact or fuzzy matching against canonical forms of scientific names in 170 distinct data sources. The Canonical form (cf) is the simplest, most complete and unambiguous form of a name. The Canonical form of scientific names consists of the genus and species – when applied – with no authorship, rank, nomenclatural annotation or subgenus.

Our system used three of the six types of matching offered by the GNR resolver: (i) exact matching; (ii) exact matching of canonical forms – this process reduce a given name to its canonical form and checks it with an exact match; (iii) fuzzy matching of canonical forms – uses a modified version of the TaxaMatch algorithm [ Rees 2008 ] and it intends to work around misspellings errors. It does a fuzzy match of the canonical form of a given name – even with mistakes – against spellings considered correct. The GNR resolver reports the matching quality (“confidence score”) for each match.

The matching module of the system is still a work in progress, but we already have obtained some relevant results to show the viability of our approach. From the LIS knowledge base we collected 7 distinct morphological descriptions: genus Varanus; species Varanus gouldii, Varanus timorensis, Varanus auffenbergi and Varanus scalaris; species groups Varanus indicus, Varanus prasinus, Varanus salvator; and Autralian spinytailed monitor lizards. Through Xper2 those morphological descriptions were exported to the SDD format and imported into the graph database, with no preprocessing. Figure 7(a) shows an overview of the resulting graph without labels. We can note the disconnectedness of the graph (7-partite graph). On the other hand, Figure 7(b) shows the same knowledge after employing the LSID unification. The graphs became connected. Before applying the LSID unification the graph had 74 distinct taxonomic units (TUs). After performing the LSID unification its total reduced to 44 TUs, i.e., 30 taxonomic units (40%) were recurring and were integrated in a single node.

The next step is to link equivalent characters of the same OTU, enabling integration of states of the same character. In the present stage of this research we apply a simple matching algorithm. One example of our preliminary results is presented in the diagram of Figure 8. As can be seen, our algorithm was able to unify all “nuchal scales” characters, by defining the same type to the edges. Moreover, we unified and congregated the possible states observed for this character across different description files.

6. Conclusion

Several initiatives propose to relate phenotype descriptions with ontologies to enable a semantic integration. The challenge is how to expand and revise the ontology while new descriptions were created. Tools which annotate descriptions with ontologies address them as an external artifact crafted apart, disregarding the synergy between building an ontolVaranus bogerti Varanus beccarii

Varanus prasinus Varanus komodoensis nuchal scales nuchal scales nuchal scales nuchal scales nuchal scales nuchal scales nuchal scales

Prasinus.sdd smooth, unkeeled granular to slightly keeled

strongly keeled triangular keeled, hull-shaped same size than head scales bigger than head scales

Varanus.sdd (a) Graph 7-partite (b) Connected graph ogy and using it. [Shirky 2005] emphasizes the importance of the semantics organically built by a community, where a binary categorization approach – in which a concept A “is” or “is not” part of a category B – to a probabilistic approach – in which a percentage of people relates A to B. This work contributes in this direction. Inspired by previous work, which explores latent semantics in folksonomies, this work analyzes standards to describe phenotypes to find a common denominator, which is the bases to link descriptions.

The main contribution of this work is to create the basis to exploit the latent semantics in the descriptions. The viability and the potential of our approach were tested by experiments. These experiments are the first steps to exploit a bigger latent semantics scenario. Moreover, having the capability of integrating knowledge around taxonomic units will enable, for instance, evolutionary biologists to generate new research questions, gain predictive insight or confront evolutionary hypotheses. More complete answers might be provided as new data sources are integrated.

Our representation in a graph database is aligned with the RDF [ Manola and Miller 2004 ] graph-based representation, which will be the next step to achieve the third layer. The challenge will be to map labels of character/characterstates in RDF properties/values. The unification of characters and states, as shown on this preliminary work, is a first and high relevant step for this mapping. Since several ontologies related to phenotype descriptions are in OWL, the relations discovered in our graph can subsidize a better matching of labels and concepts in OWL ontologies by confronting relations. For example, to enhance the match of a character label (in the graph database) with an OWL property, it is possible to consider the states allowed by the character, confronting them with the property range (values allowed by the property).

There are several possible ways to extend this work. One possible way is to incorporate morphological descriptions stored in other knowledge bases, e.g., MorphoBank (http://morphobank.org/ ) or Dryad (http://datadryad.org/ ). Another direction is to investigate correlations between State nodes and ontology terms.

Acknowledgment

Work partially financed by (CNPq 138197/2011-3), the Microsoft Research FAPESP Virtual Institute (NavScales project), CNPq (MuZOO Project and PRONEX-FAPESP), INCT in Web Science(CNPq 557.128/2009-9) and CAPES, as well as individual grants from CNPq.

References

Alves , H. and Santanche`, A. ( 2013 ). Folksonomized Ontology and the 3E Steps Technique to Support Ontology Evolvement . Journal of Web Semantics , 18 ( 1 ): 19 - 30 .

Balhoff , J. P. , Dahdul , W. M. , Kothari , C. R. , Lapp , H. , Lundberg , J. G. , Mabee , P. , Midford , P. E. , Westerfield , M. , and Vision , T. J. ( 2010 ). Phenex: Ontological annotation of phenotypic diversity . PLoS ONE , 5 ( 5 ): e10500 .

Bell , G. , Hey , T. , and Szalay , A. ( 2009 ). 323 ( 5919 ): 1297 - 1298 .

Ciccarelli , F. D. , Doerks , T. , Von Mering , C. , Creevey , C. J. , Snel , B. , and Bork , P. ( 2006 ). Toward automatic reconstruction of a highly resolved tree of life . Science , 311 ( 5765 ): 1283 - 1287 .

Clark , T. , Martin , S. , and Liefeld , T. ( 2004 ). Globally distributed object identification for biological knowledgebases . Briefings in bioinformatics, 5 ( 1 ): 59 - 70 .

Dahdul , W. M. , Balhoff , J. P. , Engeman , J. , Grande , T. , Hilton , E. J. , Kothari , C. , Lapp , H. , Lundberg , J. G. , Midford , P. E. , Vision , T. J., Westerfield , M. , and Mabee , P. M. ( 2010 ). Evolutionary characters, phenotypes and ontologies: Curating data from the systematic biology literature . PLoS ONE , 5 ( 5 ): e10708 .

Gibas , C. and Jambeck , P. ( 2001 ). Developing bioinformatics computer skills. O'Reilly Media , Inc.

Gkoutos , G. , Green , E. , Mallon , A.-M. , Hancock , J. , and Davidson , D. ( 2004 ). Using ontologies to describe mouse phenotypes . Genome Biology , 6 ( 1 ): R8 .

Hagedorn , G. ( 2007 ). Structuring Descriptive Data of Organisms - Requirement Analysis and Information Models . PhD thesis , Universita¨t Bayreuth,Fakulta¨t f u¨r Biologie, Chemie und Geowissenschaften.

Lenzerini , M. ( 2002 ). Data integration: A theoretical perspective . In Proceedings of the twenty-first ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems , pages 233 - 246 . ACM.

Mabee , P. M. ( 2006 ). Integrating evolution and development: the need for bioinformatics in evo-devo . BioScience , 56 ( 4 ): 301 - 309 .

Mabee , P. M. , Ashburner , M. , Cronk , Q. , Gkoutos , G. V. , Haendel , M. , Segerdell , E. , Mungall , C. , and Westerfield , M. ( 2007 ). Phenotype ontologies: the bridge between genomics and evolution . Trends in ecology & evolution, 22 ( 7 ): 345 - 350 .

Maddison , D. R. , Swofford , D. L. , and Maddison , W. P. ( 1997 ). Nexus: An extensible file format for systematic information . Systematic Biology , 46 ( 4 ): 590 - 621 .

Manola , F. and Miller , E. ( 2004 ). RDF Primer - W3C Recommendation . Technical report, W3C.

Page , R. ( 2008 ). Biodiversity informatics: the challenge of linking data and the role of shared identifiers . Briefings in Bioinformatics , 9 ( 5 ): 345 - 354 .

Parr , C. S. , Guralnick , R. , Cellinese , N. , and Page , R. D. ( 2012 ). Evolutionary informatics: unifying knowledge about the diversity of life . Trends in ecology & evolution, 27 ( 2 ): 94 - 103 .

Pimentcl , R. A. and Riggins , R. ( 1987 ). The nature of cladistic data . Cladistics , 3 ( 3 ): 201 - 209 .

Quan , D. ( 2007 ). Improving life sciences information retrieval using semantic web technology . Briefings in bioinformatics , 8 ( 3 ): 172 - 182 .

Rees , T. ( 2008 ). Taxamatch, a ”fuzzy” matching algorithm for taxon names, and potential applications in taxonomic databases . In Weitzman, A. and Belbin , L., editors, Provisional Abstracts of the 2008 Annual Conference of the Taxonomic Databases Working Group, Fremantle, Australia. Biodiversity Information Standards (TDWG) and the Missouri Botanical Garden .

Rodriguez , M. A. and Shinavier , J. ( 2010 ). Exposing multi-relational networks to singlerelational network analysis algorithms . Journal of Informetrics , 4 ( 1 ): 29 - 41 .

Ung , V. , Causse , F. , and Vignes Lebbe , R. ( 2010a ). Xper2: managing descriptive data from their collection to e-monographs.

Ung , V. , Dubus , G. , Zaragu¨ eta- Bagils , R. , and Vignes-Lebbe , R. ( 2010b ). Xper2: introducing e-taxonomy . Bioinformatics , 26 ( 5 ): 703 - 704 .

Vos , R. A. , Balhoff , J. P. , Caravas , J. A. , Holder , M. T. , Lapp , H. , Maddison , W. P. , Midford , P. E. , Priyam , A. , Sukumaran , J. , Xia , X. , et al. ( 2012 ). Nexml: rich, extensible, and verifiable representation of comparative data and metadata . Systematic Biology , 61 ( 4 ): 675 - 689 .