[This local archive copy is from the official and canonical URL, and .DOC format, ftp://smi.stanford.edu/pub/bio-ontology/OntologyExchange.doc; please refer to the canonical source document if possible.]
An Evaluation of Ontology Exchange Languages for Bioinformatics
Authors:
Robin McEntire (SB)
Peter Karp (Pangea Systems)
Neil Abernethy (InGenuity)
Frank Olken (LBNL)
Robert E. Kent (WSU)
Matt DeJongh (NetGenics)
Peter Tarczy-Hornoch (U of Washington, Seattle)
David Benton (SB)
Dhiraj Pathak (GW)
Gregg Helt (UC Berkeley)
Suzanna Lewis (UC Berkeley)
Anthony Kosky (GeneLogic)
Eric Neumann (NetGenics)
Dan Hodnett (NetGenics)
Luca Toldo (Merck KGaA)
Thodoros Topaloglou (GeneLogic)
August 1, 1999
Abstract
Ontologies are specifications of the concepts in a given field and the relationships among those concepts. The development of ontologies for molecular-biology information and the sharing of those ontologies within the bioinformatics community are central problems in bioinformatics. If the bioinformatics community is to share ontologies effectively, ontologies must be exchanged in a form that uses standardized syntax and semantics. This paper reports on an effort among the authors to evaluate a number of alternative ontology-exchange languages, and to recommend one or more languages for use within the larger bioinformatics community. The study selected a set of candidate languages, and defined a set of capabilities that the ideal ontology-exchange language should satisfy. The study scored the languages according to the degree to which they provided each capability. In addition, the authors performed several ontology-exchange experiments with the two languages that received the highest scores: OML and Ontolingua. The result of those experiments, and the main conclusions of this study, was that the frame-based semantic model of Ontolingua is preferable to the conceptual graph model of OML, but that the XML-based syntax of OML is preferable to the Lisp-based syntax of Ontolingua.
Ontologies, as specifications of the concepts in a given field and the relationships among those concepts, provide insight into the nature of information produced by that field and are an essential ingredient for any attempts to arrive at a shared understanding of concepts in a field. Thus the development of ontologies for molecular-biology information and the sharing of those ontologies within the bioinformatics community are central problems in bioinformatics.
If the bioinformatics community is to share ontologies effectively, the ontologies must be exchanged in some standardized form, such as using a file with a well-defined syntax and semantics. Exchange of bioinformatics ontologies will be simplified if the community can agree on a relatively small number of such exchange forms --- ideally, on one form.
This paper reports on an effort among the authors to evaluate a number of alternative ontology-exchange languages, and to recommend one or more languages for use within the larger bioinformatics community. The evaluation effort involved three separate meetings in 1998 and 1999 by the authors, as well as experiments with the proposed ontology languages. In phase I of the evaluation, the authors selected a set of candidate languages, and a set of capabilities that the ideal ontology-exchange language should satisfy.
The authors then scored the languages according to the degree to which they provided each capability. In phase II of the evaluation, the authors performed several ontology-exchange experiments with the two languages that rated the highest during phase I, which were OML and Ontolingua.
This paper describes the evaluation process and its results in more detail.
A web site maintained by the authors can be found at http://www-smi.stanford.edu/projects/bio-ontology/.
This section discusses the motivations for this work in more detail.
Ontology development is important because every biological database employs an ontology, either implicitly or explicitly, to model its data. The more fine-grained the ontology, the more precisely the database will be able to model the nuances of the data that it tries to capture. A coarse-grained ontology will model only superficial aspects of the data, and therefore may not capture data elements that are important for some problem-solving task. For example, a genome-sequence database that fails to record which genetic code is used to encode a given DNA sequence does not provide the information that users of the database will need to reliably translate each DNA sequence into the corresponding protein sequence. A semantically malformed ontology is one that incorrectly models the semantics of its application domain, and therefore yields a database whose structure corrupts or restricts the information that it is intended to hold. For example, a metabolic database that defines a one-to-one relationships between enzymes and the reactions they catalyze cannot reliably model the fact that a bifunctional enzyme catalyzes two separate reactions.
Ontology sharing is important for a number of reasons. First, ontology development is time consuming. Different bioinformatics groups who wish to develop ontologies for the same types of biological information will often arrive at a solution faster by adopting an existing ontology than by developing a new ontology de novo. For example, a group that wishes to define an ontology for microarray gene-expression data will almost certainly accomplish this task more quickly by consulting one or more existing microarray ontologies.
Second, if different bioinformatics databases that cover the same types of data (e.g., protein sequences) employ the same ontology, they simplify the problem of database integration, i.e., of processing queries across multiple biological databases. Different ontologies for the same types of data produce a semantic mismatch that complicates the multidatabase query problem.
Third, bioinformatics databases must make their schemas available to their user communities if the users are to have a full understanding of the semantics of these databases.
Fourth, ontology sharing is important because ontologies themselves constitute a form of biological knowledge that is quite valuable when shared within the bioinformatics community. For example, the taxonomy of enzymatic reactions developed by the Enzyme Commission {EC}, and the taxonomy of gene function developed by Riley {RileyOntol} are valuable bioinformatics ontologies.
Fifth, differences between ontologies puporting to represent the same biological process may lead to important insights into ways of improving those representations, and/or new insights into the underlying biology.
Ontologies are defined in the literature in a number of ways with varying degrees of formality. One prevailing definition of an ontology is a specification of a conceptualization that is designed for reuse across multiple applications. By conceptualization, we mean a set of concepts, relations, objects, and constraints that define some domain of interest.
One can argue at length about what is and is not an ontology {Gruber,Guarino}. Our view is that ontologies exist at several levels of complexity:
The GeneClinics experiment (see www.geneclinics.org) illustrates this range of complexity among different ontologies. One of the first steps of the experiment was to augment the object-oriented schema with a richer set of capabilities including disjunction, role restriction and other constraints. In the GeneClinics object database much of this information was in fact represented in the Java software interacting with the database but was hidden from the end user.
In this section we will discuss the candidate ontology-exchange languages that were evaluated by the authors. We discuss the reasons each language was selected for consideration as a bioinformatics ontology exchange language, we list the developers of each language and the reasons for its development, and provide references for each language.
The Ontolingua language was developed by a group at Stanford University for the exchange of ontologies, and was originally funded by the DARPA Knowledge Sharing Effort (Ref). Ontolingua is one of the most significant efforts to come out of the knowledge representation community and is based on the Knowledge Interchange Format (KIF), a language specifically built for the sharing of knowledge among different knowledge representation systems. The authors believed that any evaluation of languages for the exchange of ontologies must include this project.
The semantics of Ontolingua are based on the frame knowledge representation systems developed by knowledge-representation researchers {Fikes,KarpReview}.
Cyc is perhaps the best-known of the knowledge representation systems and is significant in its scope and its longevity. Cyc was developed by Doug Lenat at MCC but has since spun-off as a commercial entity, Cycorp. The underlying representation language for Cyc is called CycL, which derives from first-order predicate calculus but with extensions for additional expressivity. Cyc is one of the most significant commercial products, if not the most significant, in the marketplace currently. For this reason, as well as it's significance within the knowledge representation community and the rich expressive abilities, it was selected for evaluation.
Ontology Markup Language/Conceptual Knowledge Markup Language (OML/CKML) is a relatively new effort coming out of Washington State University that is attempting to base a system for the expression of ontologies on an XML-based syntax. The OML effort was begun in the 1990's and, though relatively young and untested, the authors believed it to have a significant representational power. This representational power combined with the interoperable nature of an XML-based language was believed to be a combination worth investigating. In addition, since OML/CKML is currently under development there is a potential for co-development to allow the bioinformatics community to influence features and expressive power of the language. There is, though, a possible disadvantage in that the language may evolve in ways that are not to the advantage of the community or is perhaps not stable or standardized.
OPM was interesting to the authors as a candidate language for exchange of ontologies because of the significance of the OPM system, a product from GeneLogic used in a number of Pharmaceutical/BioTech organizations. OPM, as a product, is used for the integration of multiple information sources, and uses an underlying object-oriented federated schema for this purpose.
Extensible Markup Language/ Resource Description Format (XML/RDF) were developed by the W3C. The current standard for the XML Schema Language is controlled by the XML Schema Working Group of the W3C. (RDF) is intended to encode metadata concerning web documents. XML/RDF were investigated as a part of the evaluation effort because of the significance of the web and web-based applications. It is clear that the web is rapidly becoming the primary method for the exchange of information and data, and that XML is currently the leading candidate for a generic language for the exchange of semi-structured objects. XML/RDF as is, without a higher level formalism that encompasses the expressivity present in frame-based languages does not go far enough to allow the kind of modeling needed in the bioinformatics community.
The Unified Modeling Language (UML) provides a set of notational conventions that can be used by software application designers/developers to model their software system. UML was developed by Rational Software and is currently backed by Rational, Microsoft and the OMG. UML was selected for evaluation because it is another widely-used system for the representation of objects and their relationships.
The Open Knowledge Base Connectivity (OKBC) is an API for accessing and modifying multiple, heterogeneous knowledge bases. OKBC is not actually an ontology exchange language – it is a programmatic API. This group considered it because its knowledge model was designed to capture ontologies. The OKBC effort began as a part of the recent DARPA High Performance Knowledge Base (HPKB) program, and is the successor of Generic Frame Protocol (GFP), a frame representation system developed at the Artificial Intelligence Center at SRI. OKBC was created because it provides a uniform model that can be understood across a number of knowledge representation systems. The work on OKBC is currently being overseen by a working group lead by Richard Fikes at Stanford. Voting members in this group are; ISI, Stanford KSL, SRI International, Cycorp, SAIC and Teknowledge.
ASN.1 was included in this evaluation because of it's historical significance as an early language for the exchange of datatypes and simple objects. The ASN.1 standard was developed as part of the OSI networking stack. It has been, and still is, being used in a number of bioinformatics applications from the National Center for Biotechnology Information. ASN.1 was also used in conjunction with the Unified Medical Language System (UMLS) project at the National Library of Medicine (NLM), however, production of ASN.1 encodings of the UMLS has been discontinued because of low demand for ASN.1 by UMLS users.
The Object Definition Language (ODL) is a relatively new standard coming out of the Object Database Management Group (ODMG) in the early 1990's. ODL was selected for evaluation because it is currently a de facto standard for a common representation of objects for object-oriented databases and programming languages and so has the potential to become a standard supported widely throughout the industry. The ODMG member companies include almost all organizations in the ODBMS/ODM industry and is very closely aligned with the OMG.
The evaluation process began with the selection of known languages for expressing ontologies. Our selection process relied on an informal review of current literature and prior knowledge of participants, but, we believe, covers the most viable candidate languages for the exchange of ontologies. The languages, once selected, were then divided among the authors for evaluation.
In order to evaluate the languages in a consistent fashion the authors arrived at a set of questions over which each candidate language would be evaluated. The questions that were distributed to members of the working group is included in Appendix A. The questions were divided into the following five major categories;
The final judgement of the authors for the initial evaluation phase was guided by a matrix of the aspects of an exchange language that were considered key to it's use by members of the Bio-Ontology Consortium (http://www-smi.stanford.edu/projects/bio-ontology/) and other groups who may want to build ontologies in the area of molecular biology. This evaluation matrix is included in Appendix B.
The authors decided that there was not a single language that stood out as the only appropriate candidate for recommendation as a language for the exchange of ontologies. It was clear that representational expressiveness was not adequate in some languages, and so they were eliminated from consideration. For example, some languages were unable to encode ground facts (instance objects). Also, some languages were in part or in whole proprietary, or had a significant cost associated with them. This was considered prohibitive to the successful adoption and use of the languages and so these languages were also eliminated . It was decided that two languages, Ontolingua and OML/CKML, provided enough expressivity to warrant a more in-depth evaluation.
The second phase of the evaluation process focused on the two candidate languages that were deemed most interesting from the initial evaluation: Ontolingua and OML/CKML.
The authors decided that it would be useful to create a small model in each language in order to judge the utility and the representational richness of each language. A set of experiments were developed to perform this detailed evaluation. Three sets of experimenters were undertaken. The three experiments and their results are discussed below.
Experiment:
Peter Karp's group at Pangea Systems performed an experiment to better understand the OML language by translating the EcoCyc gene ontology into OML. The gene ontology is a taxonomy of 150 classes that classify microbial genes according to their functions, and that was developed by Dr. Monica Riley as part of the EcoCyc project.{Riley,EcoCyc}
Within EcoCyc, the ontology can be accessed at http://ecocyc.panbio.com:1555/class-subs?object=Genes The OML encoding of the ontology can be accessed at: http://ecocyc.panbio.com/~pkarp/omlgenes.txt
Results:
Our findings were that OML was able to capture most aspects of the gene ontology. However, we identified what we consider to be a number of limitations of OML during the course of this experiment.
Experiment:
Results:
Experiment:
Peter Tarczy-Hornoch at the University of Washington in collaboration with Luca Toldo and Robert Kent performed an experiment with the general goal of using the existing GeneClinics OODB model as the basis for an ontology to assess OML/CKML and Ontolingua for ontology creation/exchange. The specific goal was to develop a small representative ontology in both Ontolingua and OML/CKML that represents key clinical and molecular entities and their linkages. design of the experiment was:
Results:
Conclusions: The expressive power of the two languages is similar and more than adequate for the purposes of expressing a part of the GeneClinics data model as an ontology. OML/CKML is however theoretically more powerful being based on a conceptual graph methodology. The Ontolingua frames semantics/paradigm on the other hand may be easier to learn since it is less of a leap from object database and object programming paradigms. The LISP syntax of Ontolingua could present a challenge to many bioinformaticians and the XML syntax of OML/CKML is likely to be more intuitive. Ideally an ontology exchange language would have an easy to learn basic semantics and syntax (like XML) but be very expressive (like OML/CKML and Ontolingua). Neither language as it stands quite achieves this ideal though a more frame-based version of OML/CKML or an XML encoding of Ontolingua might come closer. Finally, for the general bioinformatics community (not versed in ontology representation) it might be helpful to create documentation and tutorials that use biological examples.
At its last meeting, the BioOntology Core Group reached the following conclusions and recommendations.
The core group reached two major decisions for the selection of a language for the exchange of ontologies for molecular biology:
The belief of the group was that the language that the bioinformatics community needs for the exchange of ontologies should be based on frame-based semantics with an XML expression. However, the group also believed that we did not have such a language before us since Ontolingua is frame-based but without an XML expression and OML does have an XML expression, but is based on conceptual graphs, not frames.
At the meeting Peter Karp presented preliminary work that he and Vinay Chaudhri, from SRI, had done on producing an XML expression based on the OKBC knowledge model, which in turn is very closely related to Ontolingua (the Ontolingua developers were also involved in the development of OKBC).
The consensus of the group was that we recommend the use of a frame-based language with an XML syntax for the exchange of ontologies, and, to that end, the group requested that Karp and Chaudhri complete their work on the XML expression of Ontolingua, so that the group could complete its evaluation of exchange languages.
Over the last two decades, the knowledge representation and object-oriented database communities have developed a number of languages that may be used for the expression of semantic database models. These languages share many elements in common, and are exemplified by the frame knowledge representation systems used in the knowledge representation community. Frame systems have been used in many different bioinformatics projects, and the authors believe that frame systems provide the necessary representational constructs to model ontologies for molecular biology. Furthermore, frame systems have a significant amount of history and use, so that they provide a stable representational paradigm.
The authors also believe that the explosion of the web and the languages associated with it simply cannot be ignored. Acceptance of an exchange language that is expressed in a Lisp syntax will be limited within the bioinformatics community, even though the underlying representational system may be identical to that expressed in a web-based language. For this reason the authors believe that an XML-based syntax must be used for a bioinformatics ontology exchange language to increase the likelihood that the language will see widespread acceptance.
In summary, the results of this evaluation suggest two directions for future work: development of an XML expression for the Ontolingua model, or adapting OML/CKML to include a frame-based semantic model.
The authors support the use of a frame-based exchange language using an XML syntax. Several researchers on the evaluation team are currently developing a specification of XML expression of Ontolingua using OKBC. A separate set of researchers on the team are pursuing a frame-based version of OML.
The exchange language evaluation team will meet again to consider the question of whether either, or both, of these efforts provide an acceptable exchange language that meets the groups requirements.
References
EC
Edwin C. Webb, "Enzyme Nomenclature, 1992: Recommendations of the
nomenclature committee of the International Union of Biochemistry and
Molecular Biology on the nomenclature and classification of enzymes",
Eur. J. Biochem., Academic Press, 1992.
Fikes
Fikes, R. and Kehler, T., "The Role of Frame-Based Representation in
Reasoning", Communications of the Association for Computing Machinery,
1985, 28(9):904-920.
Gruber
Gruber, T.R., "A translation approach to portable ontology
specifications", Knowledge Acquisition, 1993 5:199-220.
Guarino
Guarino, N. and Giaretta, P., "Ontologies and knowledge bases towards
a terminological clarification", in Towards very large knowledge
bases, IOS Press, Amsterdam, 1995, N.J.I. Mars, pp25-32.
KarpReview
Karp, P.D., "The design space of frame knowledge representation
systems", SRI International AI Center, 1992, #520, URL
ftp://www.ai.sri.com/pub/papers/ karp-freview.ps.Z.
RileyOntol
Riley, M., "Functions of the gene products of Escherichia
coli", Microbiological Reviews, 1993, 57:862-952.
The following questions were asked of each candidate language during the Phase I evaluation process.
Language Support and Standardization:
Data model/capabilities:
Querying:
Performance:
[These questions are more about ontology tools (editors, viewers, ...) than language.]
Other Issues:
[The two questions below are asking about the ability to express non-domain relevant information in the ontology, so that, for example, one could include user model information (preferences for viewers, etc) or database access information (for access to persistent instance-level information) in the domain model.]
The table below shows the evaluation of candidate languages over general information.
Property |
ASN.1 |
ODL |
Onto |
OML/ CKML |
OPM |
XML/ RDF |
UML |
Formal Syntax? |
Yes |
No |
Yes |
Yes |
Yes |
Yes |
Yes |
Translators |
No |
Yes |
Loom,IDL,KIF,CLIPS,etc |
No |
Relational,ASN.1,XML,HTML,ER |
No |
No |
Software Tools |
Parsers |
Parsers |
WWW browsers,editors,comparison tools |
No |
Yes |
XML toolkits |
Rational Rose |
Support |
?? |
yes |
WWW documentation,FAX,tutorial,support staff |
WWW grammars, WWW examples |
Documentation,training,tutorials |
WWW sites, mailing lists, books |
Formal courses, books, tutorials |
Controlling Org |
ISO |
ODMG |
Stanford U |
WSU |
GeneLogic Inc |
W3C |
OMG |
Stability |
Stable |
Stable |
Stable |
Evolving |
Stable |
Evolving |
Stable |
Users |
Yes |
OO Vendors |
WWW users |
Intel apps |
Yes, Bix and others |
WWW developers |
many parts of industry |
Bioinfo Users |
NCBI |
Yes |
SB, Stanford RoboWeb |
Yes |
PDB |
No |
SB, (probably many other pharmas) |
Developers |
?? |
OO Vendors |
Stanford |
WSU |
GeneLogic |
many, many |
Rational Rose |
The table below shows the results of evaluation over detailed properties of the representational expressiveness of candidate languages.
Property |
ASN.1 |
ODL |
Onto |
OML/ CKML |
OPM |
XML/ RDF |
UML |
Negation |
No |
No |
Yes |
Yes |
?? |
No |
No |
Conjunction |
No |
No |
Yes |
Yes |
?? |
No |
No |
Disjunction |
No |
No |
Yes |
Yes |
?? |
No |
No |
Relations |
Yes |
Yes |
Yes |
Yes |
Yes |
Yes |
Yes |
Multiple Inheritance |
No |
Yes |
Yes |
Yes |
Yes |
Yes |
No |
Inverses |
No |
Yes |
Yes |
Yes |
Yes |
No |
No |
Multi-valued slots |
Yes |
Yes |
Yes |
Yes |
Yes |
Yes |
No |
Multiple collection types |
Yes |
Yes |
No |
No |
?? |
Yes |
No |
Number restrictions |
No |
No |
Yes |
Yes |
?? |
No |
Yes |
Slot hierarchies |
No |
No |
Yes |
Yes |
?? |
No |
No |
Facets |
No |
No |
Yes |
Yes |
?? |
No |
No |
Default Values |
No |
No |
No |
Yes |
?? |
Yes |
No |
Other slot constraints |
No |
No |
Yes |
Yes |
No |
No |
No |
Primitive Datatypes |
Standard |
Standard |
Standard |
Standard |
Standard |
None |
N/A |
Data Model |
Object w/o inheritance |
Object |
Object/Logic |
Object/Logic |
Object |
SemiStructured data |
Object |
Instances and classes |
No |
No |
Yes |
Yes |
No |
Yes |
No |
Comparison of the expressive power of the ontology-exchange
languages. The meanings of the rows are:
Appendix D - Evaluation Matrix
The table below was used by the authors to evaluate the initial candidate languages after our evaluations over the questions was complete. This table shows the desired attributes of an exchange language, and how each language can be rated along those aspects. A plus sign, '+', indicates a positive. More than one plus sign indicates more significant positives. The minus sign, '-', indicates a negative evaluation of a criteria. Also, AF indicates that the language/product is free to academic organizations, and
|
Onto |
XML/RDF |
OML |
OKBC |
OPM |
CycL |
UML/XMI |
classes & instances |
+ |
+ |
+ |
+ |
- |
+ |
+ |
multiple inheritance |
+ |
+ |
+ |
+ |
+ |
+ |
+ |
constraints |
++ |
- |
++ |
+ |
+ |
+ |
+ |
defaults |
+ |
+ |
+ |
+ |
+ |
+ |
|
expressive power |
+++ |
+ |
+++ |
++ |
++ |
+++ |
+ |
tools available* |
lisp (AF) |
Java |
lisp, Java, C |
Java, C++ |
lisp, Java, C (AF) |
||
stability |
+ |
- |
+ |
+ |
+ |
+ |
- |
support |
+ |
++ |
+ |
+ |
+ |
+ |
- |
translators |
++ |
+ |
? |
+ |
+ |
KIF. Loom |
- |
many applications |
+ |
+ |
+ |
+ |
+ |
+ |
- |
open language |
+ |
+ |
+ |
+ |
+ |
+ |
+ |
simplicity: human |
good |
low |
low |
good |
good |
low |
|
simplicity: formal |
good |
good |
good |
good |
good |
||
open to collaboration |
+ |
++ |
++ |
+ |
|||
STATUS |
out |
out |
out |
out |
out |
* By "tools available" the authors mean browsers and editors for the language.