Linköping Studies in Science and Technology
Dissertation No. 1110
Aligning Biomedical Ontologies
by
He Tan
Department of Computer and Information Science
Linköpings universitet
SE-581 83 Linköping, Sweden
The amount of biomedi al information that is disseminated over the Web
in reases everyday. Thisri hresour eisusedto ndsolutionsto hallenges
a ross the life s ien es. The Semanti Web for life s ien es shows promise
for ee tivelyande iently lo ating,integrating, queryingandinferring
re-latedinformationthatisneededindailybiomedi alresear h. Oneofthekey
te hnologiesintheSemanti Web isontologies,whi hfurnish thesemanti s
of the Semanti Web. A large number of biomedi al ontologies have been
developed. Many of these ontologies ontain overlapping information, but
it isunlikely thateventually there will be one singlesetof standard
ontolo-gies to whi h everyone will onform. Therefore, appli ations often need to
dealwithmultipleoverlappingontologies,buttheheterogeneityofontologies
hampers interoperability between dierent ontologies. Aligning ontologies,
i.e. identifying relationships between dierent ontologies, aims to over ome
this problem.
Anumber ofontologyalignment systemshave been developed. Inthese
systems various te hniques and ideashave been proposedto fa ilitate
iden-ti ation of alignments between ontologies. However, there still is a range
of issues to be addressed when we have alignment problems at hand. The
work inthis thesis ontributes to three dierent aspe ts of identi ation of
high quality alignments: 1) Ontology alignment strategies and systems. We
surveyed the existing ontology alignment systems, and proposed a general
ontology alignment framework. Most existing systems an be seen as
in-stantiations of the framework. Also, we developed a system for aligning
biomedi al ontologies (SAMBO) a ording to this framework. We
imple-mented various alignment strategies in the system. 2) Evaluation of
on-tology alignment strategies. We developed and implemented the KitAMO
frameworkfor omparative evaluationof dierent alignment strategies, and
we evaluated dierent alignment strategies using the implementation. 3)
Re ommending optimal alignment strategies for dierent appli ations. We
I would like to express my deepest gratitude to my supervisor Professor
Patri k Lambrix, who has introdu ed me to resear h. Thank you for your
invaluable guidan e, dis ussions and omments during the work. I
appre- iate your en ouragement and support. Your steady, reliable and attentive
presen e provides su h a omfortable work environment. I would like to
express my appre iation to Professor Nahid Shahmehrifor providing useful
omments andpointing out important aspe tsof theresear h world.
I am thankful to my olleagues at the Laboratory for Intelligent
Infor-mationSystems(IISLAB)andattheDivisionforDatabaseandInformation
Te hniques(ADIT).
Thanks to my friends. I appre iate the times we share hallenges and
elebratejoysoflife. Lastbutnotleast,Ifeelgratefultomyparents. Thank
you foryour understanding, supportand endlesslove. Thewarmth yougive
is the thingI herish most inthe world.
We a knowledge the nan ial support of the Swedish Resear h
Coun- il (Vetenskapsrådet), the Swedish National Graduate S hool in Computer
S ien e (CUGS), and of the EU Network of Ex ellen e REWERSE (Sixth
Framework Programmeproje t506779).
Tan,He
Linköping, Sweden
Thisthesis ontains revisedversionsof thefollowing papers:
1. P. Lambrix, H. Tan, SAMBO - A System for Aligning and Merging
Biomedi al Ontologies, Journal of Web Semanti s, Spe ial issue on
semanti web for thelifes ien es, 2006,4(3):196-206.
2. H.Tan,V.Jakoniene,P.Lambrix,J.Aberg,N.Shahmehri,Alignment
of biomedi al ontologies using life s ien e literature, Pro eedings of
the International Workshop on Knowledge Dis overy in Life S ien e
Literature,LNBI3886, Singapore, 2006,1-17.
3. B. Chen, H. Tan, P. Lambrix, Stru ture-based ltering for ontology
alignment, Pro eedings of the IEEE WETICEWorkshop on Semanti
Te hnologies in Collaborative Appli ations, 2006, 364-369.
4. P.Lambrix,H.Tan,AToolforEvaluatingOntologyAlignment
Strate-gies,Journal on Data Semanti s,LNCS 4380,2007, VIII:182-202.
5. H.Tan,P.Lambrix,AMethodforRe ommendingOntologyAlignment
Related Papers
The followingare relatedresear h arti lesnot in luded inthethesis.
1. P.Lambrix, A. Edberg, C. Manis, H.Tan, Merging DAML+OIL
bio-ontologies, Pro eedings of the International Workshop on Des ription
Logi s, Rome,Italy,2003.
2. P. Lambrix, H. Tan, Merging DAML+OIL Ontologies, J. Barzdins,
A. Caplinskas, (eds.), Databases and Information Systems (Sele ted
Papers from the Sixth International Balti Conferen e DB&IS'2004),
IOSPress, 2005,249-258.
3. P.Lambrix,H.Tan,AFrameworkforAligningOntologies,Pro eedings
of the Third Workshop on Prin iples and Pra ti e of Semanti Web
Reasoning,LNCS3703, Dagstuhl, Germany, 2005,17-31.
4. P.Lambrix, H.Tan,Ontology alignment and merging, C. Burger, D.
Davidson, R. Baldo k, (eds.),Anatomy Ontologies forBioinformati s:
Prin iples and Pra ti e, Springer,2007.
5. T. Wä hter, H.Tan,A.Wobst, P.Lambrix, M. S hroeder, A
Corpus-driven Approa h for Design, Evolution and Alignment of Ontologies,
Pro eedingsoftheWinterSimulationConferen e,2006,1595-1602.
In-vited ontribution.
Other published arti les
1. P. Lambrix, H. Tan, V. Jakoniene, L. Strömbä k, Biologi al
Ontolo-gies, C. Baker, K.H. Cheung, (eds.), Semanti Web: Revolutionizing
Knowledge Dis overy in the LifeS ien es,Springer, 2007,85-99.
2. L.Strömbä k,V.Jakoniene,H.Tan,P.Lambrix,Representing,storing
anda essingmole ularintera tiondata: areviewofmodelsandtools,
Briengs in Bioinformati s,2006,7(4):331-338.
EU Network of Ex ellen e REWERSE deliverables
1. R.Ba kofen,A.Burger,A.Bus h,G.Dawelbait,F.Fages,V.Jakoniene,
P.Lambrix, K. M Leod,S.Soliman, H.Tan, S.Will,Implementation
Dawelbait, A.Doms,F.Fages, A.Hotaran,V.Jakoniene,L.Krippahl,
P.Lambrix,K.M Leod,W.Nutt,B.Olsson,M.S hroeder,A.S hroi,
S.Soliman,H.Tan,D.Tilivea,S.Will,Requirementsandspe i ation
of use ases,REWERSE Deliverable A2-D3, 2005.
3. R.Ba kofen, M. BadeaM, P.Barahona, A.Burger, G. Dawelbait, A.
Doms,F.Fages,A.Hotaran,V.Jakoniene,L.Krippahl,P.Lambrix,K.
M Leod,S.Möller, W.Nutt, B. Olsson,M.S hroeder, S.Soliman, H.
Tan,D.Tilivea,S.Will,Usageofbioinformati stoolsandidenti ation
Introdu tion ...1 Motivation ...3 Ba kground ...5 ProblemStatement ...7 Contributions ...9 Resear h Methods ...11 Paper Summaries ...11 RelatedWork ...13 Con lusion ...16 Referen es ...19
1 Motivation
In re ent de ades,information inthebiomedi al area hasbeen growing
ex-plosively: the te hnologi al revolution has brought vast quantities of data
ongenomes,enzymes,pathwaysanddiseases. Forexample,thesequen esof
genomes for humans have been ompleted [CMP03,FJTO03℄. In the
post-genomi era,thisinformationisusedtondsolutionsto hallengesa rossthe
life s ien es,forinstan e, dete ting systemi fun tionalbehaviors ofthe ell
andtheorganism[CGGG03 ℄. Biomedi alinformation isdisseminatedovera
hugenumberofheterogeneous andautonomoussour esthatareoften
avail-able ontheInternet. For instan e, 968 datasets freelyavailable on theWeb
arelistedinthe2007databaseissueofNu lei A idsResear h[Gal07 ℄.
Ee -tivelyande ientlylo ating,integrating,queryingandinferringrelateddata
andknowledgearedi ultiesthatresear hersexperien edailyinbiomedi al
resear h [Lam05℄. The Semanti Web for life s ien es [HCLS, REW℄ shows
great promise toalleviate these di ulties (e.g.[SBL06,BC07℄).
TheSemanti Webisanextension ofthe urrentWebwhere information
is given well-dened meaning, better enabling ma hines to automati ally
pro ess information [BHL01 ℄. One of the key te hnologies in the
Seman-ti Web is ontologies, whi h will furnish the semanti s for the Semanti
Web [SBH06℄. Ontologies (e.g. [Gom99 ℄) an be seen as dening the
ba-si on epts and relations of a domain of interest, as well as the rules for
ombining these on epts and relations. They leadto abetter
understand-ing of a eld and to more ee tive and e ient handling of information in
that eld. The benets of the use of ontologies in lude reuse and sharing
of knowledge a ross platforms, and improved do umentation, maintenan e,
reliability and interoperability (e.g. [JU99℄). A re ent example is that the
RoyalSo ietyofChemistry(RSC) Publishing hasde ided toannotate their
publisheds ienti literaturewiththeGeneOntology(GO)[GO ℄andthe
Se-quen e Ontology (SO) [SO℄terms, enabling more qui kandee tivesear h
and automateddis overy of relevant resear h [GORSC℄.
Alargenumberofbiomedi alontologies have been developed,andmu h
eort hasbeen madeto advan etheir use to fa ilitate handling vast
quan-tities of information inbiomedi al resear h [BS06℄. For example, the Gene
Ontology onsortium [GO℄,OpenBiomedi alOntologies(OBO)[OBO ℄,and
SNOMEDinternational [SNOMED ℄,areinternationalresear h ooperations
for development and use of biomedi al ontologies. The National Center for
Institutes ofHealth (NIH),whose goal isto supportbiomedi al resear hers
using ontologies in theirwork.
Ontologies are developed in dierent ommunities based on their own
needs. It isinevitable that manybiomedi al ontologies ontainoverlapping
information,butusuallythereisnostraightforward waytomapthe
overlap-pingregions. Forinstan e,inthedomainofanatomy,anumberofontologies
have been developed, e.g. Foundational Modelof Anatomy (FMA) [FMA℄,
theAnatomi alDi tionary fortheAdultMouse(MA)[HMC+05℄,anatomi
ategories of theMedi al Subje t Headings (MeSH) and the U.S. National
Can er Institute (NCI) thesaurus [NCI℄. All these anatomi al ontologies
haveagreatdealin ommon,butarerepresentedindierentlanguages,and
imposedierentsyntax,stru tures,semanti s,s opesandperspe tives.
Fur-thermore,itisunlikelythateventuallytherewillbeonesinglesetofstandard
ontologies to whi h everyone will onform [NRM04℄, although mu h eort
has been devotedto developing standard ontologies inthebiomedi al area,
e.g. theCommon Anatomy Referen e Ontology (CARO) [HN07℄being
de-velopedunderNCBO.Therefore,appli ationsmustoftendealwithmultiple
ontologies. Appli ationsoftenneed ontologies fromdierent areas,orwhi h
express dierent views on one area, for example, for querying data over a
rangeofsour esannotatedwithtermsfromdierentontologies. Appli ation
developersoftenneedtodevelop ustom,task-spe i andsmallerontologies
and linkthem to thestandard ontologies. Ontology buildersmay use
exist-ingontologiesasthebasisforthe reationofnewontologiesbyextendingthe
existingontologiesorby ombining knowledgefromdierentsmaller
ontolo-gies. Inall ofthese ases,interoperabilityof multiple ontologies is hindered
by their heterogeneity, whi h has be ome a riti al issue for realizing the
vision of the Semanti Web [Noy05, BLS07 ℄.
Aligning ontologies, i.e. identifying relationships between dierent
on-tologies, aims to over omethe problem. Currently, thousands of ontologies
havebeen developed,and billionsof sour esontheWebareannotated with
ontologyterms[SWG ℄. Highqualityontologyalignmentisa ornerstone for
seamless interoperability between sour esusing dierent ontologies. It is a
ne essarypre ondition for ee tively lo ating, integrating, querying and
in-ferringinformationspreadoverheterogeneoussour es. Ourwork ontributes
to solutions for e iently identifying highqualityalignments. In thisthesis
wefo uson threeissues: developing toolsfor aligningontologies,evaluating
ontologyalignment strategies, andre ommendingoptimal alignment
to the eldof the life s ien es whi h isseen to be ahuge area for potential
appli ation developmentoftheSemanti Web,theresults anbegeneralized
and areappli able to dierent areas.
2 Ba kground
2.1 Ontologies
Ontology asthe study of existen edates ba kto an ient Greek philosophy.
Ontology studies what kinds of things exist, how these things an be
in-terrelated and what they mean in reality. The term hasbeen used in very
dierent ways within information s ien e [GG95 ℄. In this thesis, we follow
anearlydenitionprovidedintheeldoftheArti ialIntelligen e[NFF91℄:
Anontologydenesthebasi on epts andrelations ofadomainofinterest,
aswellastherulesfor ombiningthese on eptsandrelations. Thedenition
givesaroughimage ofhowanontologyisbuilt. A ordingtothedenition,
an ontology in ludesnot onlythetermsthatareexpli itly dened,but also
terms that an be inferredusing rules.
Thebenetsof theuseof ontologies in lude reuseand sharing of
knowl-edgea rossplatforms,andimproveddo umentation,maintenan e,reliability
andinteroperability(e.g.[JU99 ℄). Ontologiesareusedfor ommuni ation
be-tween people and organizations by providinga ommon terminology overa
domain. They provide thebasisfor interoperability between systems. They
an beusedformakingthe ontentininformationsour esexpli itandserve
as an index to a repository of information. Additionally, they an be used
as a basis for integration of information sour es and as a query model for
information sour es. They also support learly separating domain
knowl-edgefromappli ation-based knowledgeaswellasvalidationofdatasour es.
Overall, ontologies lead to a better understanding of a eld and to more
ee tive and e ient handling ofinformation inthat eld.
Ontologies anbe lassiedintodierenttypesdependingonwhi hofthe
omponents arerepresented andthe kindof information they an represent
(e.g [SGB01 , LTJS07℄). Controlled vo abularies are a simple type of
ontol-ogy. There are essentially lists of on epts that are enumerated expli itly.
When the on epts are organized into a hierar hi al stru ture, we obtain a
taxonomy. The relations in a taxonomy an be is-a and part-of. Thesauri
are more omplex ontologies, inwhi h there arealso relations that onne t
ontolo-gies areofthesethreetypesofontologies. Othertypesofontologies aredata
models and knowledge bases. In data models axioms whi h des ribe fa ts
that are always true intopi area of the ontology areused inaddition to
on epts and relations. Knowledge bases areoften basedon a logi . In this
thesis we fo uson ontrolled vo abularies, taxonomies andthesauri.
Ontology languages allow users to represent an ontology. The main
re-quirements of ontology languages are [AH03℄: 1, a well-dened syntax, 2,
well-dened semanti s, whi h is ane essary ondition for pre isely
des rib-ing the meaning of knowledge, 3, e ient reasoning support, whi h an be
used to, e.g. he k the onsisten y of the ontology, dete t unsatisable of
on epts and remove redundan y. 4, su ient expressive power, whi h
de-termines what kinds of knowledge an be represented in the ontology, and
5, onvenien e ofexpression. There areanumberoflanguages, e.g. RDF(s)
[RDF℄, F-Logi [KLW95℄ and OWL [OWL℄. Few of the existing ontology
languages fulllallof theserequirements.
2.2 Heterogeneity of ontologies
The heterogeneity of ontologies may o ur at dierent levels (e.g. [Kle01 ,
BEF04 ℄). At language level, ontologies are represented in dierent
lan-guages, su h as RDF(s) and OWL. At terminology level, dierent namings
of terms bring out the mismat hing between ontologies, for example,
syn-onyms, homonyms and dierent en odings. At on ept level, we en ounter
the heterogeneity of ontologies due to dierent s ope, granularity and
per-spe tive. At pragmati s level, knowledge represented in ontologies may be
indierent ontext.
The fo us of our work is to over ome the heterogeneity of ontologies
on the terminology and on ept levels. Aligning, mapping and arti ulating
ontologies aretheterms oftenusedto refer tothe task, and dierent
mean-ings of the terms aregiven inthe literature. In this thesis we usealigning
ontologies. It is dened as identifying the relationships between on epts
or relations from two dierent sour e ontologies. The relationships an be
equivalen e aswellasis-a,part-ofor anyotherkindofrelation. This
deni-tiongivesapi tureofhowtospe ify orresponden ebetweentwoontologies.
Thisdenition overs the ommon meaning ofmappingontologies. Aligning
ontologies usually is to bring together a set of ontologies, and arti ulating
ontologies is sometimes dened as aligning parts of ontologies. We also
whi hisa(minimal)unionoftwoexistingontologiesbasedontheiridentied
alignments.
Translating ontology is the ommon term referring to ta kling
het-erogeneityonthe language level, whi h isdened as onverting an ontology
representedin one ontology language to another language while preserving
the semanti s. Some tools have been developed for fa ilitating this
pro- ess, e.g. Jena [Jena℄ is a widely used tool whi h supports ontologies in
RDFs, DAML+OIL [DAML℄ and OWL. In the biomedi al area OWL has
been widely a epted, e.g. in OBO and NCBO. We do not fo us on this
issue,andassumethatontologiestobealignedareinOWL.Thepragmati s
heterogeneityis relatedtotheuseof ontologies. Thereis notmu h workon
this issue.
2.3 Aligning Ontologies
Anumberofsystemsforaligningontologieshavebeendeveloped,andvarious
methods for dis overing alignments areproposed inthese systems. Mostof
the existingsystemsaresemi-automati ,i.e. auserisinvolvedinidentifying
alignments. Todete talignments, insu hasystemthere areusually
mat h-erswhi h omputesimilarityvaluesbetween termsfromdierent ontologies,
and algorithmswhi hlter out alignment suggestionsbasedon resultsfrom
mat hers. If there are several mat hers, algorithms may be available to
ombine their results. So, alignment strategies are omposed of dierent
mat hers, ombination and ltering algorithms. Further, in su h a
system there areintera tive omponents where theuser de ides alignments
based onsuggestions.
3 Problem Statement
Aligning ontologies be omesa riti alissuewhen ollaborating appli ations
use dierent ontologies. A lot of resear h is urrently dealing with this
hallenge. A number of ontology alignment systems have been developed.
Inthesesystemsvariouste hniquesandideashavebeenproposedtofa ilitate
identifyingalignmentsbetween ontologies. However, thereis stilla rangeof
issues to be addressedwhen we have alignment problemsat hand.
No alignment system is suitable for all kinds of alignment problems.
Developers may need to build an alignment system from s rat h for their
inparti ular for ontologies fromthe eldof life s ien es. However, itis not
lear how su h systems should be developed and what would be a suitable
ar hite ture for su h a system. Further, in the eld of life s ien es domain
knowledgeisavailable intheform ofbiomedi aldomain lexi ons,databases
and olle tionsof arti les. Thisknowledge ouldbeutilized for ndinghigh
qualityalignments, but not mu h work has been done on this. So, therst
problem to be ta kled isinthis thesisis:
Problem 1: There areveryfewguidelines onbuildingalignment
sys-tems.
Although most of alignment systems and strategies are developed for
their own need, manyof them an be extendedfor otherappli ations. The
evaluationand omparisonofdierentsystemsandtheiralignmentstrategies
wouldgiveustheirstrengthsandweaknesses,andvaluableinsight into their
properties. It would leadto improvement ofthe existing alignment systems
andstrategies,butalsoprovidevaluableknowledgewhentheyarereusedfor
other alignment problems. Thus the se ondproblemis that,
Problem 2: Up to now there are few omparative evaluations, and
also thereare veryfew toolsto supportevaluations.
In most appli ations where alignments between multiple ontologies are
required, domain experts want to reuse the existing alignment strategies.
It is unlikely that there is one single strategy that outperforms the others
for all dierent kinds of ontologies, therefore domain experts must hoose
appropriatestrategiesfromdierentavailablestrategies. Itisaverydi ult
task.
Problem 3: Currently, very little knowledge is available about
ex-isting alignment methods,their appli ability,and theontologies to be
aligned, andyetno toolexiststhatfa ilitates thetask,e.g. by
re om-mending appropriate alignment strategies.
Asdis ussedinse tion2,our urrentworkwillfo usonontologieswhi h
are ontrolled vo abularies, taxonomies and thesauri. We assume that
on-tologies to be aligned are in OWL, and we do not onsider the ontext of
ontologies in appli ation. Further, we fo us on semi-automati alignment
4 Contributions
Thisthesis ontributesto aligning biomedi alontologies asfollows.
4.1 Ontology alignment systems and strategies
- Asarststepindealingwithproblem1,weadaptedthemethodforthe
lassi ation ofautomati s hemamat hingapproa hesin[RB01 ℄and
used itfor lassi ation of ontology alignment approa hes. We
lassi-ed ontology alignment approa hesbased on theknowledge they use.
The lassi ationaugmentsourunderstandingofalignment strategies.
Anoverviewoftheexistingontologyalignmentsystemsandthe
strate-gies they useisgiven inpaper1.
- As the se ond step in oping with problem 1, we proposed, based on
ontribution1,ageneralframeworkforaligningontologies. The
frame-workprovidesabasisforbuildingalignmentsystems,andprovides
sup-portfor experimenting withdierent alignment omponentsand their
ombinations. Part of the framework is similar to some steps in the
alignment pro ess introdu ed in [ES04b ℄, but our framework fo uses
onanar hite tureforalignmentsystems. Mostoftheexistingsystems
an be seen as an instantiations of our framework. This ontribution
is reportedinpaper1.
- Wedeveloped SAMBO (System for Aligningand Merging Biomedi al
Ontologies) a ording to the general framework. Currently, it is the
only alignment tool targeted towards biomedi alontologies. We
stud-iedthe hara teristi softheexistingbiomedi alontologies andsought
availablesour es inthe area, whi h an be utilized for alignment. We
implemented several alignment algorithms, ltering algorithms and a
weightedsum ombination algorithm. Wedeveloped andimplemented
strategiesbasedonthetextualdes riptionsoftheterms, is-aand
part-ofhierar hies ofthe ontologies,retrievedinstan es anduseofthesauri.
In paper 2 we report on algorithms that utilize life s ien e literature
to ompute the similarity between terms. In paper 3 we propose an
approa h that applies information about the stru ture of ontologies
inthe ltering phase. All other implemented alignment strategies are
4.2 Evaluation of alignment systems and strategies
- As the rst step in dealing with problem 2, we evaluated the
dier-entalignmentstrategiesaswellastheir ombinationswithinSAMBO.
Wealso ompared SAMBOwithotherontologyalignmentsystems. In
the evaluation we fo used on the quality of the suggestions whi h is
measured using pre ision and re all. We use pre ision as itis usually
dened ininformation retrieval, i.e. thenumberof orre tsuggestions
divided by the number of suggestions. Similarly, re all is dened as
the number of orre tsuggestions dividedby thetotalnumber of
or-re t suggestions, in this ase the expe ted suggestions. Our work is
one of the two existing evaluations whi h fo us on the quality of the
suggestions. In addition, we studied the evaluation results to dete t
strengthsandweaknessesofdierentstrategiesandtheir ombinations.
The resultsof the evaluations aredis ussed inpaper 1.
- Currently, there is no tool that provides an integrated environment
for omparative evaluations on alignment strategies and their
om-binations. The paper 4 ontributes to this issue. Based on our
gen-eralalignmentframeworkandexperimentswithSAMBO,weproposed
the KitAMO framework whi h is an integrated system to
ompara-tively evaluate non-intera tive alignment omponents and their
om-binations. We implemented a prototype a ording to the framework.
With the implementation we experimented with omparative
evalua-tions ofdierent alignment strategiesandtheir ombinations interms
of the performan e and the quality of the alignment suggestions. We
alsostudied howtheevaluationresults anbeusedtodete tstrengths
and weaknessesof dierent strategies andtheir ombinations.
4.3 Re ommendation on alignment strategies
- Up to now two methods have been proposed to deal with problem
3. They require prior knowledge about the alignment problem and
existing alignment strategies. However, su h knowledgeis usually not
available whenmaking re ommendations. In our work we proposeda
methodinwhi h the prior knowledgeis not ne essary,and whi h also
minimizestheeortfromusers. Wedis ussedthemethodinthesetting
5 Resear h Methods
Survey Inthepro ess ofunderstanding theeldof ontology alignment, we
surveyedthes ienti publi ationsonontologyalignmentandarelatedarea,
s hemamat hing. We experimentedwithpubli ly available alignment tools
using a set of biomedi al ontologies. We studied their fun tionalities and
quality of alignment. The survey results were usedas a basis for the
las-si ation ofalignment te hniques and systems,and for a general alignment
framework.
Prototyping Prototyping leads to a better and deeper understanding of
resear hproblemsandtheirproposedsolutions. Wedevelopedouralignment
systembasedonthegeneralframework,andimplementeddierentalignment
strategiesinthesystem. Weperformedanumberofevaluationsand
ompar-isons on alignment strategies and systems. The experiments demonstrated
the appli abilityof thegeneral framework,and gavea better understanding
of the alignment framework. In addition, it formed a basisfor thework on
problem 2and 3.
When dealing with problem 2,we developed a prototype for evaluating
alignment strategies. We experimented withour implemented strategies to
demonstrate and evaluate the appli ability of the prototype. The
exper-iments also indi ated that the prototype ould be applied to support the
re ommendation method we proposed for problem3. We implemented
sev-eral algorithmsinthe re ommendation method. We tested theappli ability
of the method and evaluated thequality of there ommendations when
dif-ferent algorithms were used. Thisexperiment ould, inits turn, be seen as
an appli ation study oftheevaluation prototype.
6 Paper Summaries
In this se tion we give short summaries of the ve papers en losed in this
thesis. In paper 1, 2, and 3 we investigate some fundamental problems of
aligning ontologies. Paper 4 deals with the omparative evaluation of the
dierent alignment strategies. In paper 5 we approa h making
Paper1: SAMBO-ASystemforAligningandMergingBiomedi al
Ontologies
Inthis paperwe present ageneral framework for aligningontologies. In
the framework the alignments between two sour e ontologies an be
deter-mined underthe supervisionofthe userbasedon thesuggestions generated
bydierent alignment strategiesas well astheir ombinations. Further, we
identifythe typesofalignment strategiesusedintheexistingalignment
sys-tems. We des ribeSAMBO, analignment systemwhi h exemplies theuse
of theframework. InSAMBO we implement several mat hers thatare
rep-resentatives of the dierent types of alignment strategies, a ltering and a
ombinationalgorithm. Wedis usstheevaluationsinwhi hwe ompare the
qualityof suggestions generated bythe dierent mat hers and their
ombi-nations with the ltering algorithm, and the amount of time they take to
generate the suggestions. We also ompare SAMBO with two other
align-ment systems with respe t to the quality of suggestions. We use several
well-known biomedi alontologies intheevaluations.
Paper 2: Alignment of biomedi al ontologies usinglife s ien e
lit-erature
Inthispaperwepresentourexperimentsonutilizinglifes ien eliterature
relatedto on eptsfromthesour eontologies. Weproposeabasi algorithm
as well asextensions that take the stru ture ofthe ontologies into a ount.
The algorithms build on the intuition that a similarity measure between
terms anbe omputed basedon the probabilitythatdo umentsabout one
term are also about the other term and vi e versa. The related literature
is retrieved from PubMed [PubMed ℄, a free digital ar hive of biomedi al
and life s ien es journal literature. We ompare the basi mat herwith its
extensions, and withother alignment strategiesimplemented inSAMBO in
terms of the quality of suggestions and the amount of time they take to
generate the suggestions. We also present the results of evaluations on the
dierent ombinationsof these mat hers withotherimplementedalignment
strategies.
Paper 3: Stru ture-based ltering for ontology alignment
Inthispaperweproposeamethodwhereweusethestru tural
informa-tion inthe ltering stage inthe alignment pro ess. The approa h is based
the re all in reases when thethresholds de reases in thethreshold ltering
algorithm. To keep the pre ision and in lude more orre tsuggestions, the
approa h augments thethreshold lteringalgorithm withthestru tural
in-formation. We applythealgorithm tolter outthesuggestionsfromseveral
mat hers implemented in SAMBO. We ompare the quality of the
sugges-tions lteredoutwithandwithoutthestru tureinformation,and also
om-pare theamount oftimefor omputation of thesuggestionsand ltering.
Paper 4: A Tool for Evaluating Ontology Alignment Strategies
Inpaper4weproposetheKitAMOframeworkfor omparativeevaluation
ofdierentontologyalignmentstrategiesandtheir ombinations. Wepresent
our urrentimplementation oftheframework. Withtheimplementation, we
illustrate how the system an be used to evaluate and ompare alignment
strategies and their ombinations in terms of performan e and quality of
the proposed alignments, and give suggestions on how the results an be
analyzed to obtaindeeperinsights into theproperties of thestrategies.
Paper5: AMethodforRe ommendingOntologyAlignment
Strate-gies
In this paper we propose a method that provides re ommendations on
alignment strategiesfor a given alignment problem. The method makes the
re ommendationsbasedontheevaluationofthedierentavailablealignment
strategies on severalsmall sele ted pie esfrom theontologies. In thepaper
we des ribe the basi steps of the method, and then illustrate and dis uss
the method in the setting of an alignment problem with two well-known
biomedi al ontologies.
7 Related Work
7.1 Ontology alignment systems and strategies
There area numberof systems devotedto aligning ontologies,su has
Chi-maera [MFRW00℄,FCA-Merge[SM01 ℄,PROMPT [NM03℄,IF-Map[YM03℄,
GLUE [DMDH03℄, QOM [ES04a℄, COMA++ [ADMR05℄, SST[ZK06℄.
Dif-ferent alignment strategies are implemented in these systems. In paper 1
we lassify the various alignment approa hes based on the knowledge they
use into strategies basedon linguisti mat hing, stru ture-based strategies,
aux-iliary information, and their ombinations. We give a brief overview of 13
systems and dis uss their alignment strategies a ording to this
lassi a-tion. In [EBB04 ℄ alignment approa hes are lassied into lo al methods
and global methods. Lo al methods are the mat hers based on internal
knowledge, whi h are ategorized into, terminologi al methods, stru tural
methods, extensional (based on instan es) methods, and semanti methods
(basedonmodels). Globalmethodsin ludemat hers omposedofseveral
lo- almethods, ombinationalgorithms, methods involvingtheontologiesasa
whole,learningmethods,anduserinput. 21systemsaredis usseda ording
to the lassi ations.
Generalalignmentframeworksareproposedinworksbytwoothergroups.
[BEF04 ℄ denes the alignment pro ess as a fun tion and dis usses various
onstraintsthat anbeappliedoninput, parametersandoutputofthe
fun -tion. However,howto ompute thefun tionisnot given.[ES04a ℄introdu es
an alignment pro ess withsix main steps: feature engineering, sele tion of
next sear h steps, similarity omputation, similarity aggregation,
interpre-tation and iteration. Feature engineering isto transform therepresentation
of ontologies into a format other steps an use. Sele tion of next sear h
steps restri tsasear hspa eof andidatealignmentsto redu etherun-time
omplexitybyredu ingthenumberof andidatesalignments. Mostexisting
alignmentsystemsusually he kall andidatealignments. Further, [ES04b ℄
gives 17 rules for the omputation of similarity. In paper 1 we propose our
general alignment framework. Part ofthe framework issimilar tothesteps,
similarity omputation, similarity aggregation, interpretation and iteration
in the pro ess des ribed in [ES04a ℄. The main goal of our framework is to
givea basisforbuildingalignment systems,andprovides supportfor
exper-imenting withdierent alignment omponentsand their ombinations.
Arepositoryofinformation onontologyalignmentsystems anbefound
on the ontologymat hing website (http://www.ontologymat hin g.org /). It
is updatedfrequently.
7.2 Evaluation of alignment systems and strategies
Currently, omparativeevaluationsofontologyalignmentsystemshavebeen
performedbyafewgroups. TheEUOntoWebproje t[OntoWeb℄evaluated
thesystemsPROMPTbasedonProtégé(withextensionAn hor-PROMPT),
Chimaera (des ribed, not evaluated), FCA-Merge and ODEMerge. This
visu-alization, but did not in lude tests on the qualityof thealignment
sugges-tions. Ourwork in ludes PROMPT, Chimaera,FOAM and SAMBO inthe
evaluations. The evaluations fo us on the quality of thealignment
sugges-tions as well as the time it takes to generate suggestions with these tools.
In addition, the interfa e of the systems are dis ussed in [LE03℄, and in
paper1 thedierent alignment algorithms andtheir ombinations with
dif-ferent ombination weights were evaluated with dierent ltering threshold
values. Sin e 2004, the Ontology Alignment Evaluation Initiative (OAEI,
http://oaei.ontologymat hing.org /) has organized yearly international
eval-uation ampaigns. The2006 ampaign has4tra ksgathering6data sets: a
omparison tra k, an expressive ontologies tra k, a dire tories and thesauri
tra k,anda onsensusworkshop. Ea h tra khasadierent evaluation
pur-pose. 10systemsparti ipatedinthe ampaignand 3ofthesystemsusedall
the datasets. The ampaignfo usedonthequalityofthealignment
sugges-tions. The results are reported at the international workshop on ontology
mat hing(OM-2006).
Further, tools for evaluations are developed in the latter two proje ts.
The Alignment API [Euz06℄ is the tool developed in the OAEI, whi h the
parti ipantsmay useinthe ampaign. The Alignment API in ludesseveral
evaluators whi h an ompute the pre ision, re all, fallout and f-measure
of an alignment result and a weighted symmetri dieren e between two
alignments. OLA [ELTV04℄ is a GUI appli ation implemented on top of
this API. We propose the KitAMO framework (paper 4) whi h provides
an integrated system for omparative evaluation and analysis of alignment
strategies andtheir ombinations.
7.3 Re ommendation on alignment strategies
Currently, very little resear h ta kles the problem of sele ting alignment
strategies that are optimal for given alignment problems. In [MJE06 ℄ it
is argued that nding appropriate alignment strategies should be based on
knowledge about the strategies and their previous use. As a rst step a
number offa tors (relatedto input, output, approa h,usage, ost and
do -umentation) were identied that are relevant when sele ting an alignment
strategy. The relevant data is olle ted by questionnaires. This method
requiresmu heortfrommanydierentusers. TheAnalyti Hierar hy
Pro- ess isusedto dete t suitablealignment approa hes. In[ESS05℄,APFEL, a
APFEL a set of feature parameters are de lared for the sour e ontologies,
thesimilarityassessment,andthe dierentmat hers, ombinationandlter
algorithms. To generate training data, an existing parametrization is used
and alignment suggestions aregenerated. Thesesuggestions need tobe
val-idated by the user. A ma hine learning approa h is then used to learn an
optimal parametrization. The main di ulty in this approa h is hoosing
thetrainingdata. Thevalidationofdata alsorequireseortfromusers that
have good knowledge about thearea of the ontologies. Inpaper 5 we have
proposedandtestedamethodthatisbasedontheassumptionthatstrategies
thatworkwellforaligningsele tedsmallsegmentsoftheontologieswillalso
work well for aligning the whole ontologies. The re ommendation is made
basedon twowell-known measurements ofqualityof alignment suggestions.
8 Con lusion
8.1 Summary
Aligning ontologies aims to over ome the heterogeneity of ontologies. High
quality ontology alignment is riti al for seamless interoperability between
sour es using dierent ontologies. The goal of our work isto nd solutions
for e iently identifying high quality alignments between ontologies. This
thesis ontributes tothreeproblems inthe eld: 1,development ofontology
alignment strategiesand systems,2,evaluationof alignment strategies, and
3,re ommendingsuitablealignmentstrategiesforgivenalignmentproblems.
Mu h resear h has joined the eld of aligning ontologies. On the
pub-li ation list of the ontology mat hing website, there are 14 onferen e and
journal papers in 2003, and in2006 there are 48 papers. A number of v
ar-ious alignment systems and methods have been developed. In this thesis
we lassiedontologyalignmentstrategiesbasedontheknowledgetheyuse.
The lassi ationleadstoaneasywaytounderstand theeld. Weproposed
a general framework for aligning ontologies. The framework an provide a
basis for building alignment systems, and support for experimenting with
dierent alignment strategies. Most of theexisting systems an be seen as
an instantiations of our framework. Based on the framework we developed
SAMBO, a system for aligning biomedi al ontologies. In the system we
developed dierent alignment strategies targeted at biomedi alontologies.
To improve the work in the eld, omparative evaluation of dierent
eval-uated dierent strategies implemented inSAMBO, and ompared SAMBO
with other systems. The evaluation experiments led to KitAMO, a
proto-type for omparative evaluations of alignment strategies. We implemented
the prototype,andevaluatedits appli ability. However,therearestill many
open problems in this issue, su h asevaluation riteria and onstru tion of
test ases.
Theout omeof omparativeevaluation anprovidevaluableinputto
an-other hallenge, re ommendingsuitablealignment strategiesforgiven
align-ment problems. Currently, there is very little resear h on the problem. In
this thesis we proposed a method for re ommendation of alignment
strate-gies. We implemented dierent algorithms for the dierent steps in the
method, demonstrated the feasibility of the method, and evaluated the
al-gorithms.
8.2 Future Work
In the futurewe will ontinue our workfor thethree problems asfollows.
Evaluationofalignmentstrategiesandsystems Wehaveproposedthe
KitAMOframeworkfor omparativeevaluationsofthenon-intera tive
align-ment omponents, and implemented a prototype of KitAMO. The urrent
implementation supports evaluation of alignment algorithms. In thefuture
wewillextendthesystemto supportevaluationof ombinationandltering
methods. Currently, evaluation inKitAMO an be based on the quality of
alignmentsuggestionsandthetimethatalignmentalgorithmsneedto
gener-atesuggestions. Thequalityofsuggestionsismeasured usingthere alland
pre ision. Sin e the user may need other evaluation riteria and
measure-ment methods for dierent evaluationpurposes,KitAMO will in lude more
ommonly usedevaluationapproa hes, e.g. falloutand f-measure, and may
allowusers to plug-intheir own evaluationstrategiesbasedon agiven API.
The urrenttest asesaresmallpie esfromlargebiomedi alontologies,and
they are taxonomies. We would like to build a testbed whi h in ludes test
ases withontologies ofdierent kindsandsizes, supportingdierent
evalu-ation purposes. Further,thevisualization omponentsof thesystem an be
improved.
Re ommendationof alignmentstrategies Wehaveproposedamethod
Even thoughwehaveshownthefeasibilityofourmethodandhaveobtained
good results for the alignment problem in the experiment, it is ne essary
to perform more experiments with dierent kinds of ontologies. We also
wanttoinvestigatetheinuen eofthedierent hoi esinthedierentsteps
of the method. This in ludes investigating dierent segment pair sele tion
algorithms, re ommendation measures and re ommendation algorithms. It
will alsobe interesting to lookat how theideas from[MJE06℄and [ESS05 ℄
an be usedto augment our approa h. For instan e, when more knowledge
is obtained regarding the dierent strategies and their previous use (as in
[MJE06 ℄), this knowledge ouldbe usedasarst stepto lter theavailable
strategiesandit anbeusedbythere ommendationstrategy. Alsothe
opti-mization approa h in[ESS05℄maybe usefulfor ndingbetter ombinations
aswellaswithinthere ommendationstep. Thestudieswillleadustodene
a theoreti al framework for re ommendation. Finally, we intend to develop
a toolthatsupports the framework byextending theKitAMO system.
SAMBO and ontology alignment strategies In SAMBO we have
ex-perimented with several alignment strategies. In the future we will
de-velop and implement new alignment algorithms, for example
onstraint-based algorithms, and we also want to experiment with dierent
ombina-tion andlteringapproa hes. Using KitAMO,we willevaluate thedierent
alignment strategies. The lessons learned in evaluation would lead to
im-proved strategies. On the other hand, dierent alignment strategies may
produ e new requirements on KitAMO. A lotof work has been devoted to
thenon-intera tive alignment omponents,but aligning ontologies is hardly
a fully automati task. The intera tive omponents need more study. We
have worked on integrating an intera tive ontology visualization tool into
SAMBO, whi h not only improves the user interfa e to visualize
informa-tion, but may also lead to the development of new alignment algorithms.
Further,SAMBOwillbeextendedto bettersupportontologiesfromoutside
of the biomedi al area. We also envision that after evaluation and
re om-mendation of alignment strategies, KitAMO an be asked to generate an
Referen es
[ADMR05℄ D. Aumueller, H. H. Do, S. Massmann, E. Rahm, S hema and
ontology mat hing with COMA++, Pro eedings of SIGMOD
(Demon-stration),2005.
[AH03℄ G.Antoniou,F.Harmelen,WebOntologyLanguage: OWL,S.Staab
and R. Studer (eds.) Handbook on Ontologies in Information Systems,
Springer-Verlag, 2003,76-92.
[BC07℄ Semanti Web: RevolutionizingKnowledgeDis overyintheLife
S i-en es,Baker, Christopher J.O.; Cheung, Kei-Hoi(eds.)Springer, 2007.
[BEF04℄ P. Bouquet, J. Euzenat, E. Fran oni, L. Serani, G. Stamou, S.
Tessaris, D2.2.1: Spe i ation of a ommon framework for
hara teriz-ingalignment,Te hni alreport,NoE KnowledgeWebproje t deliverable,
2004.
[BHL01℄ T.Berners-Lee, J.Hendler, O. Lassila,The Semanti Web,
S ien-ti Ameri an, 2001,284:34-43.
[BLS07℄ J.Blake,P.Lambrix,N.Shah,OntologyAlignment,Dagstuhl
Sem-inar07132, TowardsInteroperability ofBiomedi al Ontologies,Dagstuhl,
Germany, 2007.
[BS06℄ O.Bodenreider,R.Stevens,Bio-ontologies: urrenttrendsandfuture
dire tions,Briengs in Bioinformati s,2006, 7(3): 256-274.
[CGGG03℄ F.S. Collins, E. Green, A. Guttma her, M. Guyer, A vision for
the futureof genomi sresear h,Nature, 2003,422:835-847.
[CMP03℄ F.S. Collins, M. Morgan, A. Patrinos, The Human Genome
Proje t: LessonsfromLarge-S aleBiology,S ien e,2003,
300(5617):286-290.
[DAML℄ DAML+OIL.http://www.w3.org/Submission/2001/12/.
[DMDH03℄ A. Doan, J. Madhavan, P. Domingos, A. Halevy, Ontology
mat hing: A ma hine learning approa h, S. Staaband R. Studer (eds.),
HandbookonOntologiesinInformationSystems,Springer,2003,397-416.
[EBB04℄ J. Euzenat, J. Barrasa, P. Bouquet, R. Dieng, M. Ehrig, M.
H. Stu kens hmidt, P. Shvaiko, S. Tessaris, S. van A ker, I. Zaihrayeu,
T. L. Ba h, D2.2.3: State of the art on ontology alignment, Te hni al
report, NoE Knowledge Web proje t deliverable, 2004.
[ELTV04℄ J. Euzenat, D. Loup, M. Touzani, P. Valt hev, Ontology
align-ment with OLA, Pro eedings of the 3rd international Workshop on the
Evaluationof Ontology-based Tools,2004.
[ES04a℄ M.Ehrig, S.Staab,QOM- qui kontologymapping,Pro eedings of
3rd InternationalSemanti WebConferen e, LNCS3298,2004, 683-697.
[ES04b℄ M. Ehrig, Y. Sure, Ontology Mapping: An Integrated Approa h,
Pro eedings of 1st European Semanti Web Symposium, LNCS 3053,
2004,76-91.
[ESS05℄ M. Ehrig, S. Staab, Y. Sure, Bootstrapping Ontology Alignment
Methods with APFEL, Pro eedings of the International Semanti Web
Conferen e, 2005,186-200.
[Euz06℄ J. Euzenat, An API for ontology alignment, version 2.1, Ontology
Alignment Evaluation Initiative,2006.
[FaCT℄ FaCT.http://www. s.man.a .uk/
∼
horro ks/FaCT/.[FJTO03℄ M.E. Frazier, G.M. Johnson, D.G. Thomassen, C.E. Oliver,
A. Patrinos, Realizing the Potential of the Genome Revolution: The
Genomesto Life Program,S ien e, 2003,300(5617):290-293.
[FMA℄ Foundational Model of Anatomy.
http://sig.biostr.washington.edu/proje ts/fm/.
[Gal07℄ M. Y.Galperin, The Mole ular Biology DatabaseColle tion: 2007
update, Nu lei A ids Resear h, 2007, 35(Database issue):D3-D4.
[GG95℄ N.Guarino,P.Giaretta,OntologiesandKnowledgeBases: Towards
aTerminologi al Clari ation,N.Mars (ed.) TowardsVery Large
Knowl-edge Bases: Knowledge Building and Knowledge Sharing, IOS Press,
1995,25-32.
[GO℄ The Gene Ontology Consortium, Gene Ontology: tool for
the uni ation of biology, Nature Geneti s, 2000, 25(1):25-29.
[Gom99℄ A.Gómez-Pérez, Ontologi al Engineering: Astate oftheArt,
Ex-pert Update, 1999,2(3):33-43.
[GORSC℄ http://www.rs .org/Publishing/Journal s/ b/Vol ume/2 007/1
/GO-RSC.asp
[HCLS℄ W3CSemanti Web Health Care andLifeS ien es Interest Group.
http://www.w3.org/2001/sw/h ls/
[HMC+05℄ T. Hayamizu, M. Mangan, J. Corradi, J. Kadin and M.
Ring-wald, The Adult Mouse Anatomi al Di tionary: a tool for annotating
andintegrating data, Genome Biology,2005, 6(3):R29.
[HN07℄ M.A. Haendel, F. Neuhaus, D.S. Osumi-Sutherland, P.M. Mabee,
J.L.V. Mejino, C.J. Mungall, and B. Smith, CARO - The Common
Anatomy Referen e Ontology, hapter in A. Burger, D. Davidson, R.
Baldo k, (eds), Anatomy Ontologies for Bioinformati s: Prin iples and
Pra ti e, Springer, 2007.To appear.
[Jena℄ Jena - A Semanti Web Framework for Java,
http://jena.sour eforge.net/
[JU99℄ R.Jasper,M.Us hold,AFrameworkforUnderstandingand
Classify-ingOntologyAppli ations, Pro eedings of the 12thWorkshop on
Knowl-edge A quisition, Modeling andManagement, Ban,Canada, 1999.
[Kle01℄ M.Klein,CombiningandRelatingOntologies: AnAnalysisof
Prob-lems and Solutions, IJCAI Workshop on Ontologies and Information
Sharing,2001.
[KLW95℄ M. Kifer, G. Lausen, J. Wu, Logi al Foundations of
Obje t-Oriented and Frame-Based Languages, Journal of the ACM, 1995,
(42):741-843.
[Lam05℄ P. Lambrix, Towards a Semanti Web for Bioinformati s using
Ontology-basedAnnotation, Pro eedings of the 14th IEEE International
Workshops on Enabling Te hnologies: Infrastru tures for Collaborative
Enterprises,2005, 3-7.Invitedtalk.
[LE03℄ P. Lambrix, A. Edberg, Evaluation of ontology merging tools in
bioinformati s, Pro eedings of the Pa i Symposium on Bio omputing,
[LT05℄ P. Lambrix, H. Tan, Merging DAML+OIL Ontologies, J. Barzdins
andA.Caplinskas,(eds.),Databases andInformationSystems-Sele ted
Papers fromthe Sixth InternationalBalti Conferen e on Databases and
InformationSystems, IOSPress,2005, 249-258.
[LTJS07℄ P. Lambrix, H. Tan, V. Jakoniene, L. Strömbä k, Biologi al
On-tologies, C.Baker, K.H. Cheung, (eds), Semanti Web: Revolutionizing
Knowledge Dis overyin the Life S ien es, 2007,85-99.
[MeSH℄ Medi al Subje tHeadings. http://www.nlm.nih.gov/mesh/
[MFRW00℄ D. M Guinness, R. Fikes, J. Ri e, S. Wilder, An Environment
for Merging and Testing Large Ontologies, Pro eedings of the Seventh
InternationalConferen eon Prin iplesofKnowledgeRepresentation and
Reasoning,2000, 483-493.
[MJE06℄ M.Mo hol,A.Jentzs h,J.Euzenat,ApplyinganAnalyti Method
forMat hingApproa hSele tion,Pro eedingsoftheInternational
Work-shopon Ontology Mat hing, 2006.
[NCBO℄ The National Center for Biomedi al Ontology.
http://www.bioontology.org/.
[NCI℄ TheNCIthesaurus. http://n i b.n i.nih.gov/infrastru ture/ a
ore_-overview/vo abulary.
[NFF91℄ R. Ne hes, R. Fikes, T. Finin, T. Gruber, R. Patil, T. Senator,
W.R.Swartout,Enabling Te hnology for Knowledge Sharing,AI
Maga-zine,1991,12(3):36-56.
[NM03℄ N.F. Noy, M.A. Musen, The PROMPT Suite: Intera tive Tools
For Ontology Merging And Mapping, International Journal of
Human-ComputerStudies, 2003,59(6):983-1024.
[Noy05℄ N.Noy,OrderfromChaos,ACMQueue Magazine,Spe ialIssueon
Semi-Stru tured Data,2005,3(8):42-49.
[NRM04℄ N.F. Noy, D.L.Rubin, M.A. Musen. Making Biomedi al
Ontolo-gies and Ontology Repositories Work, IEEE Intelligent Systems, 2004,
19(6):78-81.
[OntoWeb℄ OntoWeb Consortium, Deliverables 1.3: A survey on ontology
tools, and 1.4: A survey on methodologies for developing, maintaining,
evaluatingand reengineeringontologies, 2002.
[OWL℄ OWL. http://www.w3.org/2004/OWL/.
[Pellet℄ Pellet. http://pellet.owldl. om/.
[PubMed℄ PubMed.http://www.n bi.nlm.nih.gov/entrez/ .
[Ra er℄ Ra er.http://www. s. on ordia. a/
∼
haarslev/ra er/ .[RB01℄ E. Rahm, P. A. Bernstein, A survey of approa hes to automati
s hemamat hing, The VLDB Journal, 2001,10:334-350.
[REW℄ REWERSE EU Network of Ex ellen e on Reasoning on the Web
with Rules and Semanti s, Deliverables of the A2 Working Group.
http://www.rewerse.net.
[RDF℄ RDF.http://www.w3.org/RDF/.
[SBH06℄ N. Shadbolt, T. Berners-Lee, W. Hall, The Semanti Web
Revis-ited,IEEE IntelligentSystems, 2006,21(3):96-101.
[SBL06℄ R. Stevens, O. Bodenreider, Y. A. Lussier, Session introdu tion:
Semanti Webs for Life S ien es, Pro eedings of Pa i Symposium on
Bio omputing, 2006,11:112-115.
[SGB01℄ R. Stevens, C. Goble, S. Be hhofer, Ontology-based knowledge
representation for bioinformati s, Briengs in Bioinformati s, 2001,
1(4):398-414.
[SM01℄ G.Stumme,A.Mäd he,FCA-Merge: Bottom-upmergingof
ontolo-gies,Pro eedings of the17th InternationalJoint Conferen e onArti ial
Intelligen e,2001, 225-230.
[SNOMED℄ SNOMED-CT. http://www.nhsia.nhs.uk/snomed/pages.
[SO℄ The Sequen e Ontology.http://www.sequen eontology.org/.
[SOFG℄ SOFG, Standards and Ontologies for Fun tional Genomi s.
http://www.sofg.org/.
[YM03℄ Y. Kalfoglou, M. S horlemmer, IF-Map: an ontology mapping
method based on information ow theory, Journal on data semanti s,
2003,I:98-127.
[ZK06℄ P. Ziegler, C. Kiefer, C. Sturm, K. R. Dittri h, A. Bernstein,
De-te tingSimilarities inOntologies withthe SOQA-SimPa kToolkit,
Pro- eedings of InternationalConferen e on Extending Database Te hnology,