=Paper=
{{Paper
|id=None
|storemode=property
|title=Checking Class Labels against Naming Conventions: First experiences with the OntoCheck
Protégé plugin
|pdfUrl=https://ceur-ws.org/Vol-897/session2-paper10.pdf
|volume=Vol-897
|dblpUrl=https://dblp.org/rec/conf/icbo/SchoberSB12
}}
==Checking Class Labels against Naming Conventions: First experiences with the OntoCheck
Protégé plugin==
https://ceur-ws.org/Vol-897/session2-paper10.pdf
Checking Class Labels against Naming Conventions:
First experience with the OntoCheck Protégé plugin
Daniel Schober1*, Vojtěch Svátek2, Martin Boeker1
1
Institute of Medical Biometry and Medical Informatics (IMBI), University Medical Center, 79104 Freiburg, Germany
2
University of Economics, Prague, Nám. W. Churchilla 4, 130 67 Praha 3, Czech Republic
ABSTRACT foster communication in collaborations with exter-
Background: Although ontology naming conventions have been nal projects to ensure effective maintenance of
proposed by policy makers, the lack of tool support for testing and
enforcing naming practices has hindered widespread compliance. modularity and orthogonality, and
We have developed OntoCheck, a Protégé plugin, which allows avoid errors and increase precision and recall in au-
testing labels in an ontology on naming inconsistencies. tomatic ontology matching and alignment algo-
Objective: We report on initial experience in applying the tool in
different settings and show that OntoCheck contributes to quality
rithms that rely on lexical/string-based similarity
assurance in a test-set of ontologies. measures.
Methods: We apply OntoCheck in four different ontology engi- To promote the application and verification of naming con-
neering efforts and test a variety of different ontologies on preva- ventions, we complemented the Protégé 4 Editor1 with tool
lence of naming issues. For each, we analyze the percentages of
class names and labels violating outlined conventions and correlate
support, the OntoCheck plugin2, extending its curation abili-
the check types to the set of OBO Foundry naming conventions. ties to help cleaning up an ontology with regard to labeling
Results: Application of OntoCheck revealed that heterogeneity inconsistencies. Besides metadata completeness checks, its
in class labels is still a common feature, even in release versions of main capabilities target the comparison of class names and
ontologies, and that many of these could be detected and rectified
by tool support. Nearly half of the OBO Foundry naming conventions
labels against self-defined or stored typographical and lexi-
could be assisted by OntoCheck, the remaining fraction relying on cal naming patterns. Detected violations can be corrected to
more complicated parsing and availability of lexica. Besides re- foster consistency in entity naming within an artifact or be-
quirements drawn from naming conventions themselves, mismatch- tween import-dependency structures.
es in string-based ontology alignment algorithms are identified as
sanity check on the impact of labelling consistency. Analysing the
Within this paper, we summarize first experiences in apply-
prevalence of false positive and negative ontology alignment mis- ing OntoCheck in a variety of practical use cases and differ-
matches could prove valuable in deriving new naming conventions ent ontology engineering efforts. OntoChecks functionalities
and test their effects in cross ontology harmonization efforts. are compared to the requirements of the OBO Foundry set
Conclusion: Our results show that typographical and syntactical
labelling heterogeneity can be improved by tool support. The appli-
of naming conventions. We provide an outlook on future
cation of OntoCheck supports the verification of naming conventions strategies to justify naming conventions and verify require-
and will ultimately ease string based ontology alignment. ments for tool support.
Our main intention is to report initial findings, testing the
1 INTRODUCTION tool on a variety of OWL ontologies and briefly reporting
Although term labeling guidelines have recently made it on the prevalence of labeling issues and naming convention
into the ‘Ten Commandments of Ontological Engineering’ violations found in the tested ontologies, as well as discuss
(Jansen & Schulz, 2011), and years after the introduction of potential future tool enhancements.
ontology class naming conventions (NC) (Schober et al.,
2009) by the OBO Foundry (Smith et al. 2009), typographic 2 METHODS
and lexical variance still persists to be a potential source for 2.1 Requirement Collection: Ontology Matching
heterogeneity in and between ontologies. But consistent
naming is not a mere aesthetic requirement, as it has been In order to draw real-life examples of synonym variance
shown to across ontologies, we surveyed string-based alignment
increase introspection of the intended meaning at mismatches found in the Ontology Alignment Evaluation
data annotation time, Initiative (OAEI)3. Of the 18 matching algorithms, we
increase readability within ontology class hierar-
chies, 1
The Protégé Ontology Editor and Knowledge Acquisition System:
http://protege.stanford.edu/, last accessed 20.01.2012
2
The OntoCheck Plugin: http://www.imbi.uni-
freiburg.de/ontology/OntoCheck/, last accessed 20.01.2012
* To whom correspondence should be addressed: schober@imbi.uni- 3
Ontology Alignment Evaluation Initiative - OAEI-2011 Campaign:
freiburg.de http://oaei.ontologymatching.org/2011/, last accessed 20.01.2012
1
Schober et al.
choose the three best, namely AgreementMaker (Cruz et al., Vehicle Sales Ontology11: A vocabulary with descriptors
2009), LogMap (Jiminez-Ruiz & Cuenca, 2011) and CODI for cars, boats, bikes, and other vehicles, serving e-
(Noessner & Niepert, 2010), and looked at the exploited commerce as complement to the GoodRelations ontology
labels and the labeling problems that lead to mapping mis- when applied in the respective field.
matches (false positives) or undetected matches (false nega- Aneurist Ontology12: An ontology providing terminologi-
tives). In addition the algorithm developers were asked via cal services for an integrated IT infrastructure for the neuro-
email to report on string mismatch examples. For each of logical research and clinical care of intracranial aneuryisms.
these, we investigated if, and which naming conventions
2.3 OntoCheck Application
would have helped avoiding those mismatches and whether
these could have been detected and curated via OntoCheck. The OntoCheck plugin13 was applied to test and curate the
selected ontologies within the Protégé 4.1 framework.
2.2 Checked Ontologies For each ontology, we created, stored and applied a different
Six ontologies were selected to be checked for labeling is- set of checks. These were either self-employed in alignment
sues via OntoCheck. Each author tested two ontologies from to the specific requirements of the particular artifact, or
different engineering efforts, namely the DebugIT project4, were taken from the respective design principle documenta-
BioTop5, GoodOD6, Aneurist7 and PatOMat8. The projects tions. Absolute counts and the percentages of found classes
cover a wide thematic scope, i.e. from the biomedical do- violating the checks were measured. Found labeling incon-
main over the educational domain up to the business do- sistencies were rectified directly or submitted to the respec-
main. Inclusion criteria for the ontologies were that they had tive curators for later amendment. The outcome of this anal-
more than forty classes, were freely accessible in OWL and ysis has been collated in Tab. 2-4. A more elaborated list
covered a wide range of domains and modeling-background and supplemental material can be found on our webpage 14.
philosophies. We here briefly describe the ontologies and
2.4 Tested Naming Conventions
their attitude to the naming conventions.
Biotop9: This biomedical upper level ontology follows the After checking the ontologies on conformance to their own
OBO Foundry conventions, but uses semantic, instead of respective naming practices, we investigated whether On-
numeric IDs. toCheck can help to enforce the 16 published naming con-
DCO (Schober et al., 2010): This large ontology serves the ventions of the OBO Foundry (Schober et al., 2009). This
semantic interoperability platform for the DebugIT project test was done by mapping the ontology-specific tests onto
on antibiotics resistance prevention. It adheres to the OBO their respective equivalents within the Foundry. These
Foundry naming conventions amended with more detailed served as a proxy to test each numbered convention, i.e. if
explicit naming conventions outlined in a design principle OntoCheck could be used to detect violations of this con-
document. It uses semantic IDs instead of numeric ones. vention type. We tracked the reason, where conventions
NTDO10: This tropical disease and epidemiology ontology could not be supported by the tool in its present state.
adheres to the OBO Foundry naming conventions, but uses
semantic instead of numeric IDs. 3 RESULTS
GoodRelations (Hepp, 2008): A vocabulary for publishing 3.1 Mismatch Examples drawn from OAEI
product details and services on the web, suitable for search
Although the complete table of mismatch pairs drawn from
engines and mobile applications in the e-commerce context.
the Ontology Alignment Evaluation Initiative (OAEI) can
GoodRelations is a small ontology (~40 classes), with so-
be found on our website, we here list a few examples, to-
phisticated design, e.g. use of longer and shortcut relational
gether with proposals for naming conventions expected to
paths impacting class naming.
alleviate the mismatches (Tab.1). The naming conventions
from the last column which can be tested with OntoCheck
are described in Section 3.3.
4
DebugIT: http://www.debugit.eu/, last accessed 20.01.2012
5
BioTop A Top-Domain Ontology for the Life Sciences: Reference Label Ontol- Label Ontology Mismatch Reason NC
http://www.imbi.uni-freiburg.de/ontology/biotop/, last accessed 20.01.2012 ogy A B
6
The GoodOD Project, http://www.iph.uni-rostock.de/Good-Ontology-
Design.902.0.html, last accessed 20.01.2012
7
@neurist – Integrated Biomedical Informatics for the Manage-ment of 11
Vehicle Sales Ontology: http://www.heppnetz.de/ontologies/vso/ns, last
Cerebral Aneurisms: http://www.imbi.uni-freiburg.de/aneurist/ontology/ 12
@neurist – Integrated Biomedical Informatics for the Manage-ment of
8
The PatOMat Project: http://patomat.vse.cz/index.html, last accessed Cerebral Aneurisms: http://www.imbi.uni-freiburg.de/aneurist/ontology/,
20.01.2012 last accessed 20.01.2012
9 13
As above: BioTop http://www.imbi.uni-freiburg.de/ontology/biotop The OntoCheck Plugin: http://www.imbi.uni-
10
NTDO – Neglected Tropical Disease Ontology: freiburg.de/ontology/OntoCheck/, last accessed 20.01.2012
14
http://www.cin.ufpe.br/~ntdo/, last accessed 20.01.2012 As above
2
First experience with the OntoCheck Protégé plugin
Bodenrei- MA:tendon NCI:Tendon One refers to bone, 1.2, 3.2 Table 2. Extract of launched checks on DCO, illustrating On-
der, one to muscle tis- toCheck’s capabilities and showing the amount of detected viola-
2005 sue tions.
Bodenrei- MA: cervical NCI: C1 Verte- Unresolved acronym 3.4, 2.1
der vertebra 1 bra 3.3 Correlating OntoCheck with OBO Foundry
2005 Naming Conventions
Svab, Associat- Chair One is a person role,
2008 edChair one is an object
1.2, 3.2 Here we list preliminary findings in correlating On-
Cruz, MA: prostate NCI:gallbladder Refer to different 3.1 toCheck’s capabilities in verifying OBO Foundry naming
2009 gland smooth mus- muscles conventions (see Tab. 3 and 4). Of the 16 conventions pub-
smooth cle tissue lished in (Schober et al., 2009), seven could be checked
muscle with our plugin, so nearly half of the Foundry conventions
Cruz, FMA:Trapezoi NCI:Trapezoid First refers to bone, 1.2, 3.2
were supported by our tool. For each naming convention,
2009 d latter to tissue
we here list aspects served by the OntoCheck tool (original
Jiminez- Review Reviewer First is paper type, 3.2 list numbering skipped where OntoCheck is not applicable):
Ruiz, second a person 1.1 Use explicit and concise names: Apply RegExp check
2011 role
for stopword detection, apply name length checks, i.e. labels
Table 1. Selected mismatches and suitable OBO Foundry naming shorter than three characters are an important source of
conventions (NC) with potential for rectification and better align- mismatches in alignment algorithms (Burgun & Bodenrei-
ment precision. der, 2005).
1.2 Use context independent names: Apply RegExp check
on explicit pre-, in-, or postfixes. E.g. all ‘ValueRegion’
3.2 Evaluation on Ontology Checks subclasses should contain either the postfix ‘ValueRegion’
Overall 61 checks were carried out on 6 different ontologies or ‘Region’, testing for the RegExp:
(the result table is available on our website). 29 of 61 checks .*ValueRegion|.*Region
(47 %) were done on ontologies without imports, either be- 1.3 Avoid taboo words: Apply RegExp check to warn on
cause the ontologies were self-sufficient (Biotop, GoodRela- ‘metalevel’ postfixes like ‘class’, ‘type’, ‘concept’, and ‘en-
tions), or to avoid redundancy, i.e. on Biotop, which is nor- tity’.
mally imported into DCO. A test that needed to be carried 2.2 Avoid conjunctions: Apply RegExp to warn on logical
out on the full import closure was the check on pre- and connectives like Boolean operators ‘and’, ‘or’. E.g. Biotop
postfixes of certain classes, as the supernode needs to be had CarbohydrateMoleculeOrResidue and OligoOrPolymer.
selected for all, e.g. biotop:Role classes. 2.4 Use positive names: Apply RegExp check for lexical
Most checks were carried out on rdf:ID and rdfs:label, but indicators of negations, e.g. checking ‘non’, ‘anti ‘or ‘dis’.
sometimes proprietary, or Dublin Core annotation properties 3.3 Use space as word separator: Apply word delimiter
were checked. checks.
For only two checks, a specific entry node was selected in 3.4 Expand abbreviations and acronyms: Apply RegExp
order to check for standard postfixes and keep the label ex- check like ‘\.’. Also a CaseConventionTest on all upper case
plicit. These subtrees were biotop:ValueRegion and bio- can detect acronyms.
top:Role (Tab. 2) in our case, but could be expanded to test 4.1 Prefer lower case beginnings: Apply word case check,
further subtrees for consistent postfix usage. e.g. CamelCase for IDs and all lower case for labels.
The remaining OBO Foundry conventions, which the tool
Tested Entity Entry Node Check Violations
abs (%)
was not explicitly able to check for were: 1.4 Avoid encod-
ing administrative metadata in names, 2.1 Use univocous
Thing CamelCase 34 (8) names and avoid homonyms, 2.3 Prefer singular nominal
Thing SpaceDelimiter 7 (4) form, 2.5 Avoid catch-all terms, 3.1 Recycle strings, 3.2 Use
Role RegExp,’Role’ postfix 2 (3) genus-differentia style names, 3.5 Expand special symbols
ValueRegion RegExp,’ValueRegion’ 167 (54) to words, 4.2 Avoid character formatting.
postfix The above would need a more thorough lexical analysis,
Thing MinCard.=1 184 (12)
requiring a lexicon, or synonym exploitation, which is not
, Thing NameEqualsLabel 304 (21) yet implemented in this version of OntoCheck. Checks on
Thing MinCard.>2 238 (40) these conventions would also require the comparison of
Thing MaxCharCount <20 3 (.5) lexical parts between different classes.
3
Schober et al.
Ontology #Checks NC Checks % NC (times) from the anatomy domain15. As lexical string mapping is the
most relevant technique in these alignment approaches
DCO 11 9 81 1.2 (4x), 2.1, 2.2, 3.1, 4.1, (Massmann et al., 2011), applying naming conventions and
4.2 enforcing them via OntoCheck should increase precision
NTDO 2 1 50 1.3 and recall of string based matching algorithms (see Tab. 1).
Biotop 13 4 30.7 1.3 (2x), 4.1, 3.1 It would be interesting to investigate which of the string
Aneurist 13 5 38.4 1.3 (2x), 3.1, 4.1
based alignment methods compared in the OAEI effort
GoodRel 11 4 36.3 2.2 (2x), 2.4, 4.1
Vehicles 10 6 60 2.2 (3x), 4.1 would profit most from a particular convention, also with
Table 3. Overall checks done on test ontologies and the amount of regard to increased matching time. To test the effect of en-
checks that could be associated with a Foundry naming conven- forced naming conventions on the ease of alignment, align-
tion. Checks for metadata completeness, i.e. on label presence, ment coherence and velocity, an ontology should be com-
were not counted as ‘1.2 use context independent names’ here. pared for its alignment precision before vs. after OntoCheck
application.
Ideally the OntoCheck plugin would make use of the LiLA
OntoCheck Function #Checks % NC(times)
framework for the linguistic analysis of entity labels in on-
CaseConventionTest 5 8.1 4.1(5x)
tologies16, which provides an interface to various natural
CompareValuesBetweenCls 1 1.6 2.1 language processing tools and resources. The LiLA API
CompareValuesForSingleCls 5 8.1 3.1(3x) (Ritze et al., 2010) is still in early development, but it would
WordDelimiterCheck 5 8.1 3.3(5x) be interesting to use it to expand OntoChecks ‘lexical
RegExp, infix 13 21.3 2.2(7x), 1.3(5x), 2.4 awareness’, as was demanded by the alignment community
RegExp, postfix 2 3.2 1.2(2x) earlier (Jimeno-Yepes et al., 2009). Leveraging on lexical
RegExp, length 3 4.9 1.2(3x)
background knowledge, such as parsers and part of speech
Table 4. Applied OntoCheck functions, mapped onto particular tagging, would not only allow for a much greater percentage
enforceable naming conventions. of OBO Foundry naming conventions to be checked (around
70 percent), but recommendations for better labels, as well
4 DISCUSSION as structural modifications could be issued. The conventions
Looking at the results of the checks shown in Tab 2-4, we profiting most from LiLa integration would be 2.1, 2.3, 3.1,
can summarize that the plugin was useful in detecting label- 3.2.
ing errors in practical application scenarios. Although a con- Work on lexically induced cross-product generation in the
siderable amount of the OBO Foundry naming conventions gene ontology project (Mungall et al., 2004) as well as clas-
could be tested with the help of OntoCheck, a significant sic ontology inference from text (Buitelaar et al., 2004,
fraction could not yet be supported as neither the ontological Svab-Zamazal & Svatek, 2008) illustrated that composition-
structure (subsumption hierarchy or relations), nor lexical al analysis of terms can contribute to directly infer structural
background knowledge (e.g. synonym lexica) are used at the patterns and make suggestions for the use of naming pat-
moment. In particular conventions 1.4, 2.1, 2.5 could be terns. In (Stevens et al., 2003), the authors show, how labels
served by simple inclusion of predefined lists of terms to be can be exploited to infer missing subsumptions, i.e. a ‘hepa-
checked for appearance in labels. Conventions number 2.1, rin biosynthesis’ is-a ‘glycosaminoglycan biosynthesis’, as
2.3, 3.1, 3.2 rely on deeper structural comparison of labels ‘heparin’ is-a ‘glycosaminoglycan’. Such inferences could
between classes, whereas 3.5 and 4.2 could be implemented only be drawn by more thorough lexical analysis given nam-
by applying standard transliteration lists, mapping special ing conventions are applied consistently. Then, by exploit-
characters onto expanded UTF codes. ing re-occurring strings among sibling classes, a ‘mor-
Bad class naming has been identified as potential source for pheme-frequency analyzer’ could infer, check or correlate
mismatches in lexical ontology alignment approaches subclass labels to a parent class affix form. E.g. if in a sub-
(Euzenat et al., 2004). The reason is that alignment plat- tree variances like X-itis, X-inflammation and inflamma-
forms such as AgreementMaker (Cruz et al., 2009) and tion-of-X occur, a tool could issue suggestions for harmoni-
PROMPT (Noy & Musen, 2001) use string distance metrics zation, i.e. by suggesting the morpheme with the highest
to discover semantic mappings between ontology classes usage frequency or the morpheme used in the common su-
(Shvaiko & Euzenat, 2008). perclass.
Within the Ontology Alignment Evaluation Initiative -
OAEI-2011 (Euzenat et al., 2011), only less than half of the
15
tools generated acceptable results trying to match classes Results for OAEI 2011:
http://oaei.ontologymatching.org/2011/results/anatomy/index.html
16
LiLA (Linguistic Label Analysis) framework for the linguistic analysis
of phrases that can occur as class or property labels in ontologies:
http://code.google.com/p/lila-project/, last accessed 20.01.2012
4
First experience with the OntoCheck Protégé plugin
Generally, a complete pre-release check specification with a Symposium on Semantic Mining in Biomedicine, SMBM 2005, EBI,
report could be generated, e.g. checking a complete set of Hinxton, UK, CEUR vol. 148
conventions from a policy provider like the OBO Foundry. Cruz IF, Antonelli FP, and Stroe C (2009). AgreementMaker: Efficient
At the moment this is hindered by the fact that in most cases Matching for Large Real-World Schemas and Ontologies. PVLDB
2(2):1586–1589
naming conventions are not outlined formally. To this end,
Euzenat, J, et al. (2011). Ontology Alignment Evaluation Initiative: six
we have recently joined forces with the ontology design
years of experience. IN: J Data Semantics XV, Lecture Notes in Com-
pattern community17 in order to formalize traceable naming puter Science, vol. 6720, 158-192
patterns. Formalizing e.g. the Foundry naming conventions, Euzenat, J, et al. (2004). State of the art on ontology alignment. Deliverable
and making them available under OntologyDesignPat- 2.2.3
terns.org would then allow a user to select a complete set of Hepp M: GoodRelations (2008). An Ontology for Describing Products and
conventions, e.g. complying with the Foundry or other suit- Services Offers on the Web. In: EKAW '08, Proceedings of the 16th in-
able policy makers. ternational conference on Knowledge Engineering: Practice and Pat-
terns. Springer, LNCS 5268, 329-346.
5 CONCLUSION Jansen L, Schulz S (2011). The Ten Commandments of Ontological Engi-
neering, Proceedings of the 3rd Workshop of Ontologies in Biomedicine
Our OntoCheck-facilitated analysis of class labels in a range and Life Sciences (OBML), Berlin, 06.-07.10. 2011
of ontologies of different size and scope has led to the detec- Jimenez-Ruiz, E and Cuenca Grau, B (2011). LogMap: Logic-based and
tion of typographically heterogeneous, unclear, unintuitive Scalable Ontology Matching. In: L.A. (ed.) The 10th International Se-
and misleading labels. It has been shown that a considerable mantic Web Conference (ISWC). LNCS, vol.7031, pp. 273–288. Spring-
amount of labels violating either a groups proprietary own er
(intra-ontology) labeling policies, or universal naming con- Jimeno-Yepes A, Jimenez-Ruiz E, Berlanga R, Rebholz-Schuhmann D
ventions outlined by policy makers could be detected and (2010). Reuse of terminological resources for efficient ontological en-
rectified with the new Protégé OntoCheck plugin. These gineering in Life Sciences. BMC Bioinformatics.10(Suppl 10):S4. doi:
10.1186/1471-2105-10-S10-S4.
results have led to the plan to carry out a more thorough
Massmann S, Raunich S, Aumüller D, Arnold P, Rahm E (2011). Evolution
analysis on labeling issues which will be based on require-
of the COMA Match System, http://disi.unitn.it/~p2p/OM-
ments rooted particularly in ontology alignment needs. 2011/om2011_Tpaper5.pdf
We hope widespread usage of our plugin will contribute to Mungall CM et al. (2004). Obol: Integrating Language and Meaning in
making ontology class hierarchies look cleaner and render Bio-Ontologies. Comparative and Functional Genomics, 5:509-520.
artifacts more informative and robust when subjected to Noessner, J and Niepert, M (2010). CODI: Combinatorial Optimization for
ontology matching and alignment approaches that leverage Data Integration–Results for OAEI 2010. In: Proceedings of the 5th
on string similarities of class names. Ultimately, we hope Workshop on Ontology Matching. Ontology Matching, page 142
this Protégé extension will ease lexical post-processing of Noy, N and Musen, M (2001). Anchor-prompt: Using non-local context for
annotated data and hence increase overall secondary data semantic matching.In: Proc. IJCAI 2001 workshop on ontology and in-
usage by humans and computers alike. formation sharing, Seattle (WA US). 63–70
Ritze D, Völker J, Meilicke C, Svab-Zamazal O (2010). Linguistic Analy-
sis for Complex Ontology Matching. Proceedings of the ISWC work-
ACKNOWLEDGEMENTS
shop on Ontology Matching (OM)
This work was supported by the Deutsche Forschungsge- Schober D et. al. (2010). The DebugIT core ontology: semantic integration
meinschaft (DFG) grant JA 1904/2-1, SCHU 2515/1-1 of antibiotics resistance patterns. Stud Health Technol Inform.;160 (Pt
GoodOD (Good Ontology Design). Vojtěch Svátek is sup- 2):1060-4.
ported by the CSF under P202/10/1825 (PatOMat). We Schober D, Smith B, Lewis S, et al. (2009). Survey-based naming conven-
thank Ilinca Tudose for the implementation of OntoCheck. tions for use in OBO Foundry ontology development. BMC Bioinfor-
matics, Vol.10, Issue 1,
http://www.biomedcentral.com/content/pdf/1471-2105-10-125.pdf
REFERENCES
Shvaiko, P and Euzenat, J (2008). Ten challenges for ontology matching.
Bodenreider, O, Hayamizu, TF, et al.: (2005). Of mice and men: Aligning In: On the Move to Meaningful Internet Systems (OTM Conferences)
mouse and human anatomies. AMIA Annu Symp Proc. pp. 61–65, Smith B, Ashburner M, Rosse C et al. (2007). The OBO Foundry: coordi-
http://www.ncbi.nlm.nih.gov/pmc/articles/PM nated evolution of ontologies to support biomedical data integration.
Buitelaar P, Olejnik D, Sintek M (2004). A protégé plug-in for ontology Nat Biotechnol. 25(11):1251–1255. doi: 10.1038/nbt1346.
extraction from text based on linguistic analysis. Proceedings of the 1st Stevens R, Wroe C, Bechhofer S, Lord P, Rector A, Goble C (2003). Build-
European Semantic Web Symposium (ESWS). ing Ontologies in DAML + OIL, Comp Funct Genomics. February;
Burgun A and Bodenreider O (2005). An ontology of chemicals helps 4(1): 133–141.
identify dependence relations among Gene Ontology terms, In: First Svab-Zamazal O, Svatek V (2008). Analysing Ontological Structures
through Name Pattern Tracking, EKAW 2008,
17
Ontology Design Patterns: http://www.springerlink.com/content/j808w138253683m4/fulltext.pdf
http://ontologydesignpatterns.org/wiki/Main_Page, last accessed
20.01.2012
5