Towards quantitative measures in applied ontology

02/16/2012
by   Robert Hoehndorf, et al.
University of Cambridge
0

Applied ontology is a relatively new field which aims to apply theories and methods from diverse disciplines such as philosophy, cognitive science, linguistics and formal logics to perform or improve domain-specific tasks. To support the development of effective research methodologies for applied ontology, we critically discuss the question how its research results should be evaluated. We propose that results in applied ontology must be evaluated within their domain of application, based on some ontology-based task within the domain, and discuss quantitative measures which would facilitate the objective evaluation and comparison of research results in applied ontology.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

10/28/2017

An Ontology to support automated negotiation

In this work we propose an ontology to support automated negotiation in ...
07/28/2018

Ontology-Grounded Topic Modeling for Climate Science Research

In scientific disciplines where research findings have a strong impact o...
02/20/2015

Development of a VO Registry Subject Ontology using Automated Methods

We report on our initial work to automate the generation of a domain ont...
11/25/2020

The Landscape of Ontology Reuse Approaches

Ontology reuse aims to foster interoperability and facilitate knowledge ...
05/08/2020

Knowledge Patterns

This paper describes a new technique, called "knowledge patterns", for h...
03/05/2012

Development of an Ontology to Assist the Modeling of Accident Scenarii "Application on Railroad Transport "

In a world where communication and information sharing are at the heart ...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Applied ontology is an emerging discipline that applies theories and methods from diverse disciplines such as philosophy, cognitive science, linguistics and formal logics to perform or improve domain-specific tasks. Scientific disciplines require a research methodology which yields reproducible and comparable results that can be evaluated independently. Methodological progress in applied ontology will be recognized when different methods generate results that can be objectively compared, such that it can be evaluated as to whether the methods yield better results. To illustrate, text mining

, which was firmly established as a scientific discipline in the early 1990s, leveraged knowledge from computational linguistics, cognitive science, philosophy, graph theory, machine learning, and other areas of computer science to create new, large scale methods to uncover information from natural language documents. Although text mining could, in principle, be evaluated from the perspective of computational linguistics (i.e., how well a particular linguistic theory was implemented and applied), it is most commonly evaluated from the perspective of scientific contribution (i.e., how well some scientific question was addressed by the text mining method) through quantitative measures that include precision and recall based on comparison to a gold standard, the F-measure (i.e., the harmonic mean between precision and recall), the area-under-curve (AUC) in an analysis of the receiver operating characteristic (ROC) curve

[1], or the use of kappa-statistics to determine agreement between manual evaluation by domain experts. Establishment of a common set of (quantitative) measures has made it possible to compare different methods and approaches in text mining with respect to their contribution to particular tasks and has the field as a whole allowed to measure its progress.

Research in applied ontology currently lacks established quantitative metrics for evaluating its results. More importantly, applied ontology lacks agreement about the perspective from which its results should be evaluated. Evaluation of applied ontology research is more often than not based on criteria stemmed from philosophy, knowledge representation, formal logics or “common sense”, while an evaluation based on the domain of application is rarely performed [2]. In the absence of commonly agreed criteria for evaluating research results, the evaluation and comparison of research in applied ontology is prone to subjectivity, lack of transparency, opinion, preference and dogma. Furthermore, the lack of established evaluation criteria for applied ontology research hinders the development of an effective research methodology for the field of applied ontology: before a research methodology in any scientific field can be established, it is first necessary to determine what constitutes a research result, what constitutes a novel research result (i.e., what does it mean that two research results are different), and what constitutes a better result than another (i.e., how can two competing results be compared and evaluated). Only after these questions are answered will it be possible to design a research methodology in a scientific field than enables the field as a whole to make progress with respect to the evaluation criteria that the discipline has established.

Here, we being to explore ways for evaluating research in applied ontology. Our arguments and examples will primarily focus on ontologies that are used in science, in particular for biomedicine, but we believe that many of our arguments will hold for research in other areas of applied ontology as well. From a certain perspective, this work is a continuation and extension of the work of Barry Smith [3], who made one of the first moves towards questioning the status of applied ontology as a scientific discipline, and who already stated that “[c]entral to ontology (science) is the requirement that ontologies […] should be tested empirically”.

Our central claim will be that research in applied ontology must be evaluated within the domain to which it is being applied. More precisely, we claim that research results in applied ontology need to be evaluated with respect to a specific task that is supposed to be achieved, and that any contribution in applied ontology should be motivated by a task and evaluated based on it. In many cases, the results of such an evaluation can be quantified, leading to objective, empirical criteria for the success of applying ontological methods within a domain. We do not intend to propose a comprehensive list of such evaluation criteria, but rather aim to make suggestions for the research community in the hope to spawn a discussion that results in community agreement and a first step towards the development of a commonly agreed research methodology for our field.

2 Biomedical ontologies

At the end of the 1990s and early 2000s, genetics made a leap forward with the availability of the first genome sequences for several species. The availability of genome sequences for multiple species enabled comparative genomic analysis, and it was recognized that a large part of the genetic material in different species was conserved and that many of the genes in different organisms have similar functions. The Gene Ontology (GO) [4] was designed as a controlled vocabulary to provide stable names, textual definitions and identifiers to unify descriptions of functions, processes and cellular components across databases in biology. Today, with the rise of high-throughput sequencing technology, genome sequences for thousands of species are becoming available, and large international research projects, such as the 5,000 genomes project (which aims to sequence the genomes of 5,000 insects and other arthropods) or the Genomes 10k project (which aims to sequence the genomes of 10,000 vertebrate species), will collect even more data in the near future. High-throughput technologies are not limited to genome sequencing, but entered other areas in biology as well, from high-throughput phenotyping (to determine the observable characteristics of organisms, often resulting from targeted mutations) over microarray experiments (to determine gene expression) to high-throughput screening (in drug discovery). The amount of data produced in biology today makes the design of strategies for integration of data across databases, methods for retrieving the data and developing query languages and interfaces a central and important part of research in biology. The prime purpose of ontologies such as the GO is to address these challenges that arose in biology and bio-medicine within the last few years.

To facilitate the integration of databases, retrieval of data and the provision of query languages, ontologies provided not only terms and textual definitions, but also a basic structure. Initially, this structure was not expressed in a formal language (i.e., a language with an explicit semantics). Instead, ontologies were seen as graph structures in which nodes represent terms and edges relations (such as is-a or part-of) between them. Reasoning over these graphs was stated as operations on the graph, in particular the composition of edges and the transitive closure [4]. It was not until much later that formal languages were used to represent biomedical ontologies and recast the graph operation in terms of deductive inference over formal theories [5, 6, 7, 8].

The graph structure of biomedical ontologies is not only a valuable feature to improve retrieval and querying, but is widely used in the form of Gene Set Enrichment Analysis (GSEA) [9] to analyze gene expression. GSEA utilizes the graph structure of the GO to determine whether a defined set of genes shows statistically significant, concordant differences between two biological states; it utilizes the annotation of sets of genes with GO terms and the GO graph structure and inference rules to statistically test for enriched GO terms. A large number of tools were developed to perform GSEA, and it has lead to discoveries of cancer mechanisms [9], evolutionary differences in primates [10], and GSEA is now a standard tool in many biological analyses, as evidenced by more than 3,000 citations111Based on Google Scholar, 12 Jan 2012. for the original paper. Similar enrichment analyses are now being performed using ontologies of other domains, such as the Human Disease Ontology [11].

The graph structure of ontologies is also widely utilized for semantic similarity analyses [12]. Semantic similarity measures apply a metric on an ontology in order to compare the similarity between data that is annotated with classes in the ontology. Some metrics are based on the distance between two nodes in the ontologies’ graph structure, while others compare sets of classes that are closed with respect to relations in the ontology. In some cases, the metrics include further information, such as the information content that a class in an ontology has within a given domain. Importantly, however, semantic similarity measures rely on the number and the kind of distinctions that the ontology developers have made explicit, and utilize the explicit semantics in an ontology only indirectly.

Another application of ontologies is in text mining and literature search and retrieval. The availability of a common terminology throughout biology enables the task of named entity recognition, i.e., the identification of standardized terms in natural language texts. When terms from ontologies can reliable be detected in natural language texts, ontologies can be used for retrieving text documents from literature archives such as PubMed

[13]. This task is made easier when terms in ontologies are widely used, and several biomedical ontologies have been evaluated based on how well their terms can be recognized in scientific literature [14].

Neither of the applications of biomedical ontologies we discussed so far actually relies on formalized semantics, axioms, the use of knowledge representation languages and automated reasoning, or philosophical foundations. Nevertheless, the past years have seen a rapid increase in applying formal methods to biomedical ontologies. In particular, the Web Ontology Language (OWL)

[15] is now widely used to represent biomedical ontologies [5]. In some cases, more expressive languages such as first- and monadic second-order logic is used to specify ontologies, in particular for biological sequences [16] and molecular structures and graphs [17]. The stated aims of using the axiomatic method [18] and knowledge representation languages for biomedical ontologies are manifold, and include, among others, the search for philosophical rigour and a foundation in “good” philosophy [19], providing “unambiguous” documentation of the meaning of terms in a vocabulary [20], verifying the consistency of a (conceptual) data model [21, 22], verifying the consistency of data with respect to a data model [21, 23], enabling complex retrieval and querying through automated reasoning [24], integrating multiple ontologies [25, 26], and decreasing the cost of developing and maintaining an ontology [27, 28]. Furthermore, the application of formal methods in biomedical ontologies has the potential to reveal mistakes in the design of ontologies and thereby improve their utility for scientific analyses [7, 29]. Several projects have started to axiomatize biomedical ontologies [22, 26, 30, 31], and these projects have led to changes in the ontologies and the detection and removal of contradictory statements [25, 22]. Other researchers have suggested changes to improve ontologies’ structures and axioms based on applying formal, ontological and philosophical methods [7, 29, 32, 33], or they provide ontological interpretations of domain-specific knowledge by applying some formal ontological theory to some phenomena in a domain [34, 32, 35, 36].

Despite the large number of research projects that apply formal ontological theories to (scientific) domains, no common evaluation criteria are being applied in these studies. Examples of criteria of evaluation include formal consistency [37], identification of unsatisfiable classes [22, 25], conformance to a “good” philosophy (i.e., some particular philosophical view) [38, 39, 19], user acceptance [40], conformance to naming conventions [41] or the recall of ontology class labels in scientific literature [14]. Only few of these criteria actually evaluate the application of ontologies to some task, while the majority of these criteria evaluate the research results based on philosophical, formal and technical criteria that lie within the domain of ontology or its underlying technologies themselves.

3 Towards quantitative evaluation criteria for research results in applied ontology

The selection and application of evaluation criteria may provide us with the means to distinguish research in applied ontology from research in non-applied ontology. In applied ontology, ontologies are being used for some task within a domain, and that task lies usually outside of the domain of ontology itself222

A notable exception to this is when we apply ontological methods to the domain of ontology itself, and classify different kinds of ontology, analyze the types of relations between classes, relations, instances and individuals, etc. Such an ontology could, for example, be used to provide the conceptual foundation of an ontology editor, to enable interoperability between different ontology learning algorithms, in portals providing access to different ontologies, or in an ontology evaluation framework.

. Consequently, quality criteria for research in applied ontology will be derived from the task to which the research is being applied, and not from the domain of ontology itself. On the other hand, the search for philosophical foundation and rigour, including the demonstration that a particular philosophical theory is capable of expressing distinctions that are being made within a domain, are examples of research goals of non-applied ontology, not of applied ontology, because the aim of the research and its evaluation will generally lie within the realm of ontology or philosophy, not within the domain of application. Applying a particular philosophical theory can, in many cases, improve the utility of an ontology within a domain. Nevertheless, the fact that a philosophical theory can be applied within a domain alone does not, in our opinion, constitute a result in applied ontology; on the other hand, that the application of a particular philosophical theory or perspective improves the utility of an ontology for some task in a domain would constitute a result in applied ontology.

We can also observe who or what directly benefits from a particular aim of research in ontology: either the users and uses of an ontology, ontology-based applications, and specific tasks to which ontologies are being applied, or the developers and maintainers of an ontology. Developers and maintainers of ontologies will benefit directly from decreased maintenance work, ease of construction and the availability of technical documentation, while users and applications of an ontology will only benefit indirectly from such research goals (and these benefits would normally have to be demonstrated). Users and applications of ontologies benefit from the community agreement which ontologies can bring about and their resulting potential for ontology-based data annotation and integration, retrieval and querying, novel scientific analyses, and in some cases consistency verification of data. In particular, users and uses of ontologies will benefit from something that ontologies can do, and research in applied ontology – ontology research to serve some domain’s use case – will have to be measured on how well they perform their task.

One of the most widely cited applications of ontologies in science is their potential to facilitate community agreement of the meaning of terms in a domain. These terms are frequently used as metadata in scientific databases and publications. Consequently, applying ontologies to standardize the vocabulary used as meta-data can enable the integration and interoperability of databases and research results. Yet, how could such a research result – an ontology that is intended to effectively standardize the meaning of terms in a vocabulary in order to support interoperability and integration – be evaluated? Since the prime aim of such a research result is to achieve community agreement, an obvious evaluation criterion would be to conduct a user-study that evaluates whether different users can consistently apply terms within a standardized task such as the annotation of a data set with classes from an ontology. For this task, Kappa statistics can be applied and a value can be reported that measures the degree to which annotator can consistently apply an ontology within the task [42]. Alternatively, an integrated scientific analysis of the data in multiple databases between which interoperability is intended to achieve can be performed and evaluated on a scientific use case. For example, the development of formal definitions for phenotype ontologies [43] could be quantitatively evaluated by using these definitions to integrate multiple model organism databases and analyze the integrated knowledge with regard to its potential for revealing novel candidate genes for diseases [44].

The support of queries and the accurate retrieval of data is another task that ontologies or the axioms in ontologies are developed for. Information retrieval is a discipline in computer science for which rigorous quantitative evaluation criteria are available [45], often based on the comparison to a gold standard or a set of positive and negative examples based on which statistical measures can be applied. Quantitative measures include the F-measure (the harmonic mean between precision and recall) or the area-under-curve in an analysis of the receiver operating characteristic (ROC) curve [1]. If an ontology, or axioms in an ontology, are intended for retrieval, measures of this kind can be applied to demonstrate the success.

In many cases, axioms in ontologies are added in order to enable novel queries that make distinctions which could not be made before. For example, adding axioms that assert a partonomy to a purely taxonomic representation of anatomical structures enables new kind of queries based on the use of parthood relations. Such a result – the addition of new axioms in order to enable novel types of queries and retrieval operations – can be evaluated using the same quantitative measures as ontology-based retrieval. All of these descriptions assume that there is already some data which is being retrieved using queries over the ontology. In the absence of such data, e.g., when a new ontology is proposed within a domain with the intent to use this ontology to annotate data in the future, data could still be simulated and used in the evaluation.

Further applications of formalized ontologies include the verification of data with respect to certain constraints that are expressed within the ontology. For example, in the domain of biological pathways333A biological pathway is a series of interactions that lead to a particular outcome, such as a chemical product or the realization of a particular function., the BioPax ontology [21] has been proposed, and one of its aims is to verify pathway data with respect to the model that the BioPax ontology provides. Similarly, a recent study used formal ontological analysis and automated reasoning to investigate the consistency of data stored in the BioModels database (a database of computational models in systems biology), and identified a large number of incorrectly characterized database entries [23].

Applications of ontology research in scientific analyses and in the process of making novel scientific discoveries are maybe the best evaluated contributions in applied ontology, since the contributions that ontology research can make in these areas is commonly subject to the same evaluation criteria as other contributions in the scientific domain of application. For example, the GSEA method was evaluated both using statistical measures and experimentally verified data that has been extensively studied [9], and the use of semantic similarity measures to identify interacting proteins based on similar Gene Ontology annotations is rigorously evaluated and compared using ROC analysis and correlation coefficient analysis [46]. In each case, the scientific domain to which ontology-based methods are being applied has established (and often demands) quantitative evaluation criteria that can guarantee – at least to some degree – the objective and empirical evaluation and comparison of research results.

There are several other tasks that may fall in the domain of applied ontology research. For example, formal ontological analysis can be applied to specify a (conceptual) model, verify its consistency and identify modelling choices that potentially lead to faulty results; or formal ontology can be applied to formally and “unambiguously” specify the meaning of terms in a vocabulary (e.g., to enable communication between autonomous intelligent agents). Some of these tasks can also be evaluated quantitatively: while consistency of a conceptual model is a binary quality that relies on a consistency proof, incorrect consequences can be estimated using predefined tests that aim to make inferences of a certain kind. The “unambiguous” formal specification of the meaning of a term using an ontology would require a meta-theoretical analysis and a completeness proof for the ontology.

Application Potential evaluation method Quantifiable result
Establish community agreement of the meaning of terms within a domain; facilitate data annotation; support integration and interoperability User study; integrated analysis; completeness proof Inter-annotator agreement, Cohen’s Kappa
Retrieval Comparison to gold standard, ROC analysis, unit tests area-under-curve, F-measure, precision, recall
Scientific analysis (GSEA, semantic similarity) Comparison to gold standard, statistical analysis p-value, area-under-curve, F-measure, precision, recall
Consistency of data automated reasoning, performance evaluation computational complexity
Determine the consistency of a (conceptual) model automated reasoning, consistency proof consistent or not (binary)
Test the accuracy of a (conceptual) model automated reasoning, unit tests (for inferences), unit tests (for application) number of unsatisfiable classes, number of tests passed/failed
Table 1: Opportunities for the quantitative evaluation of research results in applied ontology.

To summarize, depending on the task that is being performed using some ontology research result, we will be able to derive different quality criteria, some of which are illustrated in Table 1. However, the heterogeneity of ontology-based applications and ontology-driven approaches prevents the application of a single quality and evaluation criteria. Instead, we have to evaluate research results in applied ontology in conjunction with a particular task to which this result is being applied. For example, instead of evaluating the quality of an ontology that represents biological pathways, we evaluate with respect to different tasks that it is intended to perform. For example, may be used to achieve community agreement about the terms used to annotate pathway databases (task ), and we can evaluate with respect to this task. On the other hand, may also be used to verify the consistency of biological pathway data (task ), and we may evaluate with respect to . It may then turn out that achieves one task very well while its performance in a second task is poor.

Finally, robustness of research results in applied ontology can be evaluated based on how well a research result in applied ontology performs in multiple tasks, or how well it can be adapted to other tasks, including tasks that are performed in other domains. Robustness can be evaluated based on how much the quantitative evaluation changes under changing application conditions. For example, if the research result is an ontology that is being developed for the semantic annotation of a particular database and has been demonstrated (e.g., based on a user-study and the report of the inter-annotator agreement) to perform well in this task, changing the database and performing a similar study and quantitative evaluation allows to evaluate robustness: does the quantitative evaluation result change significantly, or does it remain the same? If the quantitative evaluation results do not change significantly under changing conditions of application (or they improve), evidence for a robust research result have been found. Notably, it is the application of quantitative evaluation criteria that enables the direct comparability of the suitability of a research result in different tasks, and therefore enables an objective demonstration of robustness.

4 Proposed evaluation and quality criteria

Several evaluation methods for research in applied ontology have been proposed, and multiple studies have attempted to evaluate the quality of ontologies in several domains. To the best of our knowledge, no study has yet emphasized the need for objective, quantitative evaluation criteria for applied ontology research; on the contrary, many criteria that aim to measure the “quality” of ontologies are derived from philosophical considerations or based on social considerations. In particular, several studies emphasize the need to treat ontologies similarly to scientific publications and propose an evaluation strategy similar to scientific peer review. For example, Obrst et al. aim to identify “meaningful, theoretically grounded units of measure in [ontology]” and perform an extensive review of previous ontology evaluation attempts, including a brief discussion of application-based evaluation approaches and quantifiable results [2]. However, Obrst et al. dismiss application-based evaluation strategies since they are “expensive to carry out”, and seem to favour evaluation by humans based on principles derived from common sense, from formal logics or from philosophy (especially in the form of philosophical realism). A similar route is being taken by Smith who suggests that peer review of ontologies should become standard practice, since “[p]eer review provides an impetus to the improvement of scientific knowledge over time” [3]. Such a peer review system is intended to be adopted by the OBO Foundry ontology community [47, 3]. The use of expert peer review to evaluate ontologies seems like an uncontroversial suggestion since peer review is the established method for evaluating contributions throughout science. However, different from applied ontology, most scientific fields have widely accepted, and in many cases objective, quantitative, criteria on which peer reviewers can base their judgement. The criteria for peer review proposed by Smith [3], Orbst et al. [2], and others [48], are largely derived from “common sense” or particular philosophical positions and have not been demonstrated to improve the performance of ontology-based research in any application. Peer review cannot be used to evaluate research results when there is no agreement as to how a discipline is supposed to achieve scientific progress and how these achievements can be measured. In the absence of accepted and empirically tested criteria, peer review will merely reflect the personal opinions of the reviewers, and not lead to a fair evaluation of a research result’s quality or its fitness for a particular purpose.

A prime example of a conflict resulting from the lack of accepted, empirically tested evaluation criteria is the realism debate [49, 50, 39]. The realism debate is an argument between the proponent of the “realist methodology”, who argue that ontologies must be evaluated with respect to some form of philosophical realism [39, 2, 48], and researchers in applied ontology who argue for a research methodology in which ontological decisions are motivated and evaluated by applications and not philosophical considerations [49, 50]. The difference between a philosophical and an application-centric perspective may be one of the reasons for misunderstandings between the two sides in this debate: while one side attacks the other in the realm of philosophy – where philosophical positions are attacked and defended, and some of the arguments have been exchanged between philosophers for thousands of years –, the proponents of an application-centric view would expect it to be a matter of empirical investigation to determine which ontological design decisions address the needs of the ontology users better than another. In many cases, it may turn our that two philosophical theories are indistinguishable for a particular scientific task (e.g., when two theories are empirically equivalent), in which case the particular choice of philosophical explanation will not affect the performance of an application: when it is in principle impossible to design an experiment that can distinguish between two alternative theories, we would leave the realm of empirical science if we attempt to defend or attack either theory.

There are some notable previous studies which applied quantitative measures for formalized ontologies in biomedical applications. For example, Boeker et al. [40] “aim to analyze the correctness of the use of logic by the OBO Foundry or close-to OBO Foundry ontologies and related mappings”, and they identify approximately 23% of the axioms in the evaluated ontologies as incorrect based on the judgement of four experts. These results are consistent with another study by Hoehndorf et al. [25] that evaluates contradictory class definitions in OBO ontologies and identifies several thousand unsatisfiable classes using automated reasoning. Common to these two studies is that they evaluate ontologies based on aspects that can be derived from their formal representations alone, assuming that considerations such as the consistency of an ontology or the absence of undesired inferences from an ontology will always give some indication about an ontology’s quality. However, for some tasks, not even consistency of an ontology is required. Boeker et al. “hypothesize that the main and only reason why [the problematic axioms have] little affected the usefulness of these ontologies up to now is due to their predominant use as controlled vocabularies rather than as computable ontologies” [40], already acknowledging that their evaluation has not addressed the main task for which the evaluated ontologies are being applied, but rather some task (retrieval through automated reasoning) that these ontologies could potentially also be used for. Similarly, Hoehndorf et al. identify several unsatisfiable classes in biomedical ontologies, but fail to identify the problems that these may cause – except again in the hypothetical task of using automated reasoning to answer queries over the ontologies. Even more problematically, the evaluated ontologies are successfully being used for automated reasoning although they contain unsatisfiable classes and may lead to undesirable inferences. In these reasoning tasks, applications such as database queries restrict the types of operations that are being performed over the ontologies. From a certain perspective, applications provide an interface to formal ontologies that may limit the ontology to a lower expressivity than the knowledge representation language in which the ontology is formulated. For example, if an inference mechanism that lacks the capability to interpret negation is used to process an ontology, retrieval operations can be successful even in the presence of contradictory class definitions or inconsistencies. Similarly, undesired inferences may disappear when only certain kinds of queries can being performed. In some cases, “incorrect” consequences may even be desirable: for example, in analyses that utilize measures of semantic similarity, the similarity between two classes in an ontology may not coincide with some ontological distinctions (such as between occurrent and continuant) that are deemed to be “correct” within the domain [2], but lead to undesired results in a similarity-based analysis.

Finally, Widely used criteria for ontology development are the OBO Foundry principles444Both accepted and proposed principles can be found on http://obofoundry.org/crit.shtml. The accepted criteria (as of 12 Feb 2012) include that ontologies must be (1) open and freely available to all users, (2) that they are expressed using a common syntax, (3) that they use unique URIs, (4) that they include versioning information, (5) that their content is clearly delineated and orthogonal to other ontologies, (6) that they contain natural language definitions for all their terms, (7) that they define relations based on patterns described in the OBO Relationship Ontology, (8) that they are well-documented and (9) have multiple, mutually independent users, (10) that they are developed collaboratively while (11) only a single person is responsible for the ontology, (12) that they follow ontology naming conventions, and (13) that they are maintained in light of scientific advance. The majority of these criteria (1-5, 9-13) are intrinsically social criteria; although their discussion is outside the scope of the current article, it must be emphasized that these criteria are highly valuable for enabling wide access to the content of the ontologies within the OBO Foundry, and therefore serve to enable scientific discourse about and investigations into the ontologies and their content. The remaining criteria could be classified based on methods to demonstrate that they are satisfied and based on the tasks which they aim to improve. For example, while the inclusion of textual definitions (criterion 6) and documentation (criterion 8) may improve comprehensibility of ontologies, comprehensibility also depends on the quality of the textual definitions and documentation; user-studies may be used to evaluate and quantify the effect of these criteria, and compare them against automated methods to generate textual definitions [51]. Criterion (7), the use of relations that are defined in the OBO Relationship Ontology [7], aims to improve interoperability between ontologies through reuse of relations. However, while relation names may be reused across ontologies, it is not always guaranteed that they are reused in the same meaning. To quantify whether criterion (7) succeeds in enabling interoperability between ontologies, it would, for example, be possible to combine two ontologies that both use relations from the OBO Relationship Ontology, and evaluate whether or not they yield desired inferences (i.e., a comparison of inferences against a gold standard).

5 Conclusions

Our central position is that research results in applied ontology should always be evaluated against a task for which they are intended, i.e., the evaluation must be based on the behavior of the whole system consisting of the ontology and the applications that are based on it. Whether the research result is an ontology, or an ontology design pattern, or a method to formulate particular phenomena within a domain, the benefit it can bring to the domain cannot be evaluated based on the research result alone; instead, any evaluation criteria must evaluate the whole system consisting of the research result and a task – or a set of tasks – to which the ontology-based research is being applied.

Many of the applications and tasks in which ontologies play a role are amenable to quantitative evaluation criteria. Quantitative measures enable the objective comparison of research results and can play a crucial role in the evaluation of research. We have reviewed several common applications of applied ontology research in biomedicine, and discussed potential quantitative evaluation measures for each of them.

These quantitative measures could be adopted in addition to already established qualitative evaluation criteria, and they can also serve to justify and refine existing qualitative measures. For example, while we have little doubt that qualitative measures such as formal consistency and the absence of contradictory statements in an ontology are useful and important quality criteria, we believe that many of these qualitative criteria can be derived from underlying quantitative measures of the performance of ontology-based research within a task: consistency of ontologies is a useful criterion because many applications of ontologies depend on consistency and because consistent ontologies will often lead to better outcomes in whatever application an ontology is being applied for.

Furthermore, with the application of quantitative measures, ontology development methodologies can be evaluated with respect to how well they ensure or improve the performance of research results in particular tasks within a domain. More importantly, accepted evaluation criteria for research results are the first step in developing a research methodology for the field of applied ontology. It was not our aim to establish such criteria for research in applied ontology; instead, we believe that we, as a community of scientists and scholars, must increase our efforts towards establishing such evaluation criteria for research in applied ontology, based on which we can derive a research methodology within our field.

Acknowledgements

We are grateful to Prof. Heinrich Herre, who contributed to this work over many years through discussions and critical remarks.

References

  • [1] Tom Fawcett. An introduction to ROC analysis. Pattern Recognition Letters, 27(8):861 – 874, 2006. ROC Analysis in Pattern Recognition.
  • [2] Leo Obrst, Werner Ceusters, Inderjeet Mani, Steve Ray, and Barry Smith. The evaluation of ontologies: Toward improved semantic interoperability. In Christopher J. O. Baker and Kei-Hoi Cheung, editors, Semantic Web: Revolutionizing Knowledge Discovery in the Life Sciences. Springer, 2007.
  • [3] Barry Smith. Ontology (science). In Proceeding of the 2008 conference on Formal Ontology in Information Systems: Proceedings of the Fifth International Conference (FOIS 2008), pages 21–35, Amsterdam, The Netherlands, The Netherlands, 2008. IOS Press.
  • [4] Michael Ashburner et al. Gene ontology: tool for the unification of biology. Nature Genetics, 25(1), May 2000.
  • [5] Ian Horrocks. OBO flat file format syntax and semantics and mapping to OWL Web Ontology Language. Technical report, University of Manchester, March 2007. http://www.cs.man.ac.uk/~horrocks/obo/.
  • [6] Christine Golbreich and Ian Horrocks. The obo to owl mapping, go to owl 1.1! In Christine Golbreich, Aditya Kalyanpur, and Bijan Parsia, editors, Proceedings of OWL: Experiences and Directions 2007 (OWLED-2007). CEUR-WS.org, 2007.
  • [7] B. Smith, W. Ceusters, B. Klagges, J. Köhler, A. Kumar, J. Lomax, C. Mungall, F. Neuhaus, A. L. Rector, and C. Rosse. Relations in biomedical ontologies. Genome Biol, 6(5):R46, 2005.
  • [8] Robert Hoehndorf, Anika Oellrich, Michel Dumontier, Janet Kelso, Dietrich Rebholz-Schuhmann, and Heinrich Herre. Relations as patterns: Bridging the gap between OBO and OWL. BMC Bioinformatics, 11(1):441+, 2010.
  • [9] Aravind Subramanian, Pablo Tamayo, Vamsi K. Mootha, Sayan Mukherjee, Benjamin L. Ebert, Michael A. Gillette, Amanda Paulovich, Scott L. Pomeroy, Todd R. Golub, Eric S. Lander, and Jill P. Mesirov. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proceedings of the National Academy of Sciences of the United States of America, 102(43):15545–15550, 2005.
  • [10] Kay Prufer, Bjoern Muetzel, Hong-Hai Do, Gunter Weiss, Philipp Khaitovich, Erhard Rahm, Svante Paabo, Michael Lachmann, and Wolfgang Enard. Func: a package for detecting significant associations between gene sets and ontological annotations. BMC Bioinformatics, 8:41+, February 2007.
  • [11] Paea LePendu, Mark Musen, and Nigam Shah. Enabling enrichment analysis with the human disease ontology. Journal of Biomedical Informatics, 2011. In press.
  • [12] Philip Resnik. Semantic similarity in a taxonomy: An Information-Based measure and its application to problems of ambiguity in natural language.

    Journal of Artificial Intelligence Research

    , 11:95–130, 1999.
  • [13] A. Doms and M. Schroeder. GoPubMed: exploring PubMed with the Gene Ontology. Nucleic Acids Res, 33(Web Server issue):783–786, Jul 2005.
  • [14] Lixia Yao, Anna Divoli, Ilya Mayzus, James A. Evans, and Andrey Rzhetsky. Benchmarking ontologies: Bigger or better? PLoS Comput Biol, 7(1):e1001055, 01 2011.
  • [15] B. Grau, I. Horrocks, B. Motik, B. Parsia, P. Patelschneider, and U. Sattler. OWL 2: The next step for OWL. Web Semantics: Science, Services and Agents on the World Wide Web, 6(4):309–322, November 2008.
  • [16] Robert Hoehndorf, Janet Kelso, and Heinrich Herre. The ontology of biological sequences. BMC bioinformatics, 10(1):377+, November 2009.
  • [17] Janna Hastings, Oliver Kutz, and Till Mossakowski. How to model the shapes of molecules? combining topology and ontology using heterogeneous specifications. In Proceedings of Deep Knowledge Representation Challenge Workshop, 2011.
  • [18] David Hilbert. Axiomatisches Denken. Mathematische Annalen, 78:405–415, 1918.
  • [19] Janna Hastings, Colin R. Batchelor, Christoph Steinbeck, and Stefan Schulz. What are chemical structures and their relations? In FOIS, pages 257–270, 2010.
  • [20] Robert Hoehndorf, Colin Batchelor, Thomas Bittner, Michel Dumontier, Karen Eilbeck, Rob Knight, Chris J. Mungall, Jane S. Richardson, Jesse Stombaugh, Eric Westhof, Craig L. Zirbel, and Neocles B. Leontis. The RNA ontology (RNAO): An ontology for integrating RNA sequence and structure data. Applied Ontology, 6(1):53–89, April 2011.
  • [21] Emek Demir et al. The BioPAX community standard for pathway data sharing. Nature Biotechnology, 28(9):935–942, September 2010.
  • [22] Christopher J. Mungall, Michael Bada, Tanya Z. Berardini, Jennifer Deegan, Amelia Ireland, Midori A. Harris, David P. Hill, and Jane Lomax. Cross-product extensions of the gene ontology. Journal of biomedical informatics, February 2010. in press.
  • [23] Robert Hoehndorf, Michel Dumontier, John H. Gennari, Sarala Wimalaratne, Bernard de Bono, Daniel L. Cook, and Georgios V. Gkoutos. Integrating systems biology models and biomedical ontologies. BMC Systems Biology, 5(1):124+, August 2011.
  • [24] Alan Ruttenberg et al. Advancing translational research with the semantic web. BMC Bioinformatics, 8(Suppl 3):S2+, 2007.
  • [25] Robert Hoehndorf, Michel Dumontier, Anika Oellrich, Dietrich Rebholz-Schuhmann, Paul N. Schofield, and Georgios V. Gkoutos. Interoperability between biomedical ontologies through relation expansion, upper-level ontologies and automatic reasoning. PLOS ONE, 6(7):e22006, July 2011.
  • [26] Christopher J. Mungall. OBO flat file format 1.4 syntax and semantics [draft]. Technical report, Lawrence Berkeley National Laboratory, 2011. http://berkeleybop.org/~cjm/obo2owl/obo-syntax.html.
  • [27] C. Goble and R. Stevens. State of the nation in data integration for bioinformatics. Journal of Biomedical Informatics, 41(5):687–693, 10 2008.
  • [28] Michael Bada, Robert Stevens, Carole Goble, Yolanda Gil, Michael Ashburner, Judith A. Blake, Michael J. Cherry, Midori Harris, and Suzanna Lewis. A short study on the success of the gene ontology. Web Semantics: Science, Services and Agents on the World Wide Web, 1(2):235–240, February 2004.
  • [29] B. Smith, J. Williams, and S. Schulze-Kremer. The ontology of the gene ontology. AMIA … Annual Symposium proceedings / AMIA Symposium. AMIA Symposium, pages 609–613, 2003.
  • [30] Christopher J. Mungall, Colin Batchelor, and Karen Eilbeck. Evolution of the sequence ontology terms and relationships. Journal of Biomedical Informatics, March 2010.
  • [31] Gary Schindelman, Jolene Fernandes, Carol Bastiani, Karen Yook, and Paul Sternberg. Worm phenotype ontology: integrating phenotype data within and beyond the c. elegans community. BMC Bioinformatics, 12(1):32, 2011.
  • [32] Stefan Schulz, Holger Stenzhorn, Martin Boeker, and Barry Smith. Strengths and limitations of formal ontologies in the biomedical domain. RECIIS – Electronic Journal in Communication, Information and Innovation in Health, 3(1):31–45, 2009.
  • [33] Robert Hoehndorf, Axel-Cyrille Ngonga Ngomo, and Janet Kelso. Applying the functional abnormality ontology pattern to anatomical functions. Journal for Biomedical Semantics, 1:4, 2010.
  • [34] Johannes Rohl and Ludger Jansen. Representing dispositions. Journal of Biomedical Semantics, 2(Suppl 4):S4, 2011.
  • [35] Stefan Schulz, Boontawee Suntisrivaraporn, Franz Baader, and Martin Boeker. SNOMED reaching its adolescence: Ontologists’ and logicians’ health check. International Journal of Medical Informatics, 78(Supplement 1):S86–S94, 2009.
  • [36] S. Schulz, H. Stenzhorn, and M. Boeker. The ontology of biological taxa. Bioinformatics, 24(13):i313, 2008.
  • [37] Oliver Kutz and Till Mossakowski. A modular consistency proof for dolce. In AAAI, 2011.
  • [38] B. Smith, W. Ceusters, and R. Temmerman. Wuesteria. Studies in Health Technology and Informatics, 116:647–652, 2005.
  • [39] Barry Smith and Werner Ceusters. Ontological realism: A methodology for coordinated evolution of scientific ontologies. Appl. Ontol., 5:139–188, August 2010.
  • [40] Martin Boeker, Ilinca Tudose, Janna Hastings, Daniel Schober, and Stefan Schulz. Unintended consequences of existential quantifications in biomedical ontologies. BMC Bioinformatics, 12(1):456, 2011.
  • [41] Daniel Schober, Barry Smith, Suzanna Lewis, Waclaw Kusnierczyk, Jane Lomax, Chris Mungall, Chris Taylor, Philippe Rocca-Serra, and Susanna-Assunta Sansone. Survey-based naming conventions for use in obo foundry ontology development. BMC Bioinformatics, 10(1):125, 2009.
  • [42] Jacob Cohen. A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20(1):37–46, April 1960.
  • [43] Christopher Mungall, Georgios Gkoutos, Cynthia Smith, Melissa Haendel, Suzanna Lewis, and Michael Ashburner. Integrating phenotype ontologies across multiple species. Genome Biology, 11(1):R2+, 2010.
  • [44] Robert Hoehndorf, Paul N. Schofield, and Georgios V. Gkoutos. Phenomenet: a whole-phenome approach to disease gene discovery. Nucleic Acids Research, 39(18):e119, 2011.
  • [45] Cornelis Joost van Rijsbergen. Information Retrieval. Butterworths, 1979.
  • [46] Tao Xu, LinFang Du, and Yan Zhou. Evaluation of GO-based functional similarity measures using s. cerevisiae protein interaction and expression profile data. BMC Bioinformatics, 9(1):472, 2008.
  • [47] Barry Smith et al. The OBO Foundry: coordinated evolution of ontologies to support biomedical data integration. Nat Biotech, 25(11):1251–1255, 2007.
  • [48] Ludger Jansen and Stefan Schulz. The ten commandments of ontological engineering. In Heinrich Herre, Robert Hoehndorf, and Frank Loebe, editors, OBML 2011 Workshop Proceedings, pages G–11. Institut fuer Medizinische Informatik, Statistik und Epidemiologie (IMISE), Markus Loeffler, 2011.
  • [49] Gary H. Merrill. Ontological realism: Methodology or misdirection? Applied Ontology, 5(2):79–108, January 2010.
  • [50] Michel Dumontier and Robert Hoehndorf. Realism for scientific ontologies. In Antony Galton and Riichiro Mizoguchi, editors, Formal Ontology in Information Systems – Proceedings of the Sixth International Conference (FOIS 2010), pages 387–399, 2010.
  • [51] Robert Stevens, James Malone, Sandra Williams, and Richard Power. Automating class definitions from OWL to English. In Proceedings of Bio-Ontologies 2010: Semantic Applications in Life Sciences SIG at the 18th Annual International Conference on Intelligent Systems for Molecular Biology (ISMB 2010), July 2010.