|
|
AMIA Annu Symp Proc. 2005; 2005: 864–868. | PMCID: PMC1560629 |
Copyright This is an Open Access article: verbatim copying and redistribution of
this article are permitted in all media for any purpose Alignment of Multiple Ontologies of Anatomy: Deriving Indirect Mappings
from Direct Mappings to a Reference Songmao Zhang, PhD and Olivier Bodenreider, MD, PhD U.S. National Library of Medicine, Bethesda, Maryland National Institutes of Health, Department of Health & Human Services, Email: smzhang/at/math.ac.cn, Email: olivier/at/nlm.nih.gov Objective To investigate the indirect alignment of two anatomical ontologies through
a reference ontology and to compare it to direct alignment between
these two ontologies. The ontologies under investigation are the Adult
Mouse Anatomical Dictionary (MA) and the NCI Thesaurus (NCI). The Foundational
Model of Anatomy serves as reference ontology. Methods The direct alignment employs a combination of lexical and structural similarity. The
indirect alignment simply derives mappings from direct alignments
to the reference ontology. Results The indirect MA-NCI alignment yielded 703 mappings and the direct alignment 715, 654 of
which are common to both. The mappings specific to one
approach were analyzed. Conclusions When a reference ontology exists, indirect alignment of multiple ontologies
through a reference represents a valid, cost-effective alternative
to pairwise alignment. Anatomical knowledge is central to biomedical applications. Various representations
of anatomy have been developed including anatomical ontologies (e.g., the
Foundational Model of Anatomy, Adult Mouse Anatomical
Dictionary) and broader ontologies covering anatomy (e.g., GALEN, NCI
Thesaurus). Despite differences in modeling principles and representation
formalisms, these ontologies are expected to be compatible with each
other. Mappings among ontologies constitute an enabling resource for
applications such as knowledge sharing and application system communication. In
particular, such mappings represent a crucial component of
the Semantic Web in which the semantic annotation of resources will
inevitably draw on multiple ontologies [ 1]. Mappings among ontologies can be built pairwise, i.e., an alignment is
created between every two ontologies. Alternatively, one ontology can
be selected as the reference for mapping. All other ontologies only need
to be mapped to this reference ontology and the pairwise mappings can
be derived from the mappings to the reference ontology. These two approaches
to aligning multiple ontologies are illustrated in . The objective of this study is to compare two approaches to aligning multiple
ontologies: pairwise ontology alignment and alignment through a
reference ontology. More precisely, we investigate the quality of a mapping
between two ontologies (O 1–O 2) automatically derived from mappings created between these two ontologies
and a reference ontology (O 1-R and O 2-R), compared to the direct mapping between O 1 and O 2. The three ontologies of anatomy under investigation in this study are: the
Foundational Model of Anatomy (FMA) 1, the Adult Mouse Anatomical Dictionary (MA) 2, and the anatomy subset of the NCI Thesaurus (NCI) 3. To our knowledge, this is the first attempt of deriving mappings automatically
among anatomical ontologies from the alignment of these ontologies
to a reference. The creation of pairwise mappings among ontologies of anatomy draws on
previous work, namely the methodology developed for aligning the Foundational
Model of Anatomy and GALEN [ 2]. For a survey of ontology alignment techniques, see for example [ 3]. The Adult Mouse Anatomical Dictionary (MA) is a structured controlled vocabulary describing the anatomical structure
of the adult mouse [ 4]. It comprises 2,404 concepts. Each concept has one name (e.g., Head muscle and Adrenal artery). Additionally, 240 concepts have a total of 259 synonyms (e.g., Limb has synonym Extremity). The ontology is represented as a directed acyclic graph whose edges
represent the relationships IS-A and PART-OF. Every concept is connected to others through IS-A or PART-OF relationships. However, about 38% of the concepts do not have
any IS-A relationship to others (e.g., Knee PART-OF Hindlimb is the only hierarchical relation available for Knee). On the other hand, nearly 4% of the concepts have more than
one IS-A relationship to others (e.g., Hand phalanx is both a kind of Phalanx and Hand digit bone). The version used in this study was downloaded on December 22, 2004 (under
the name Mus adult gross anatomy in the Open Biomedical Ontologies 4). The NCI Thesaurus (NCI) provides standard vocabularies for cancer research [ 5] and its anatomy class describes naturally occurring human biological
structures, fluids and substances. The ontology is available in
the Ontology Web Language (OWL). There are 4,410 anatomical concepts (accounting
for about 12% of all NCI concepts). Every concept
has one preferred name (e.g., Abdominal esophagus). Additionally, 1,207 concepts have a total of 2,371 synonyms (e.g., Orbit has synonym Eye socket). Except for the root ( Anatomic Structure, System, or Substance), every anatomical concept has at least one IS-A relationship to another concept, and nearly 4% of concepts have
more than one IS-A relationship to others (e.g., Radius bone is both a kind of Long bone and Bone of the upper extremity). In addition, anatomical concepts are also connected by a PART-OF relationship (named ANATOMIC STRUCTURE IS PHYSICAL PART OF). The version used in this study is version 04.09a (September 10, 2004). The Foundational Model of Anatomy (FMA) is an evolving ontology that has been under development at the University
of Washington since 1994 [ 6]. Its objective is to conceptualize the physical objects and spaces
that constitute the human body. The underlying data model for FMA
is a frame-based structure implemented with Protégé. 71,202 concepts
cover the entire range of macroscopic, microscopic and
subcel-lular canonical anatomy. In addition to preferred terms (one
for each concept), 52,713 synonyms are provided (up to 6 per concept). For
example, there is a concept named Uterine tube and its synonym is Oviduct. Because single inheritance is one of the modeling principles used in the
FMA, every concept (except for the root) stands in a unique IS-A relation to other concepts. Additionally, concepts are connected by seven
kinds of PART-OF relationships (e.g., part of, constitutional part of, regional part of). For the purpose of this study, we considered as only one PART-OF relationship and its inverse HAS-PART the various kinds of partitive relationships present in FMA. The version
used in this study was downloaded on December 2, 2004. As illustrated in , this study compares the direct alignment between MA and NCI to the indirect
alignment automatically generated from mapping both MA and NCI
to FMA, the reference ontology. Details about the three phases of our
study follow: 1) three direct alignments: MA-NCI, MA-FMA, and NCI-FMA; 2) indirect
alignment between MA and NCI through their direct alignments
with the FMA; and 3) comparison of the direct alignment MA-NCI to
the indirect alignment obtained through the FMA. Direct alignment Aligning two ontologies directly consists of comparing terms across systems
in order to identify one-toone concept matches. This first step constitutes
the lexical component of our method. Following is the identification
of structural matches. Inter-concept relationships are compared
in order to identify similar relations among lexical matches across
systems. The interested reader is referred to [ 7] for additional precisions about our method. Identifying matches lexically The lexical alignment compares two systems at the term level, by exact
match and after normalization. This process makes the source and target
terms potentially compatible by eliminating such inessential differences
as inflection, case, hyphen, and word-order variation. Both preferred
terms and synonyms in the two systems are used in the alignment process. Moreover, UMLS
synonymy is used to identify additional matches. For
example, Profunda femoris artery in MA and Deep femoral artery in NCI, although lexically different, are considered as a match because
they name the same anatomical concept in UMLS. Our method does not address
partial lexical matches. Identifying matches structurally In order to facilitate the comparison of relations across systems, the
structural alignment first consists of acquiring the inter-concept hierarchical
relationships, including IS-A and PART-OF, and their inverses. Missing inverse relations are complemented as necessary. Inference
rules are used to generate a partitive relation between
a specialized part and the whole or between a part and a more generic
whole. Finally, the relations reified in the names of some FMA concepts
are represented explicitly (e.g., <Heel, PART-OF, Foot> was augmented from <Heel, IS-A, Subdivision of foot>). Once all relations are represented consistently, the structural alignment
is applied on the matches resulting from the lexical alignment in order
to identify similar relations to other matches across systems. For
example, match concepts Forelimb in MA and Upper extremity in NCI exhibit similar relations to other matches in the two systems, including Limb (through IS-A), Arm and Hand (through HAS-PART) across systems. Such structural similarity is used as positive evidence
for the alignment. Instead of similar relations, one match may exhibit
relations to other matches in opposite directions in the two systems. Such
relations suggest a structural conflict across systems. For example, in
MA Pericardial cavity is in HAS-PART relationship to Pericardium, while in the FMA Pericardial cavity is defined as part of Pericardial sac which is part of Pericardium. These conflicts are used as negative evidence for the alignment, indicating
the semantic incompatibility between concepts across systems despite
their lexical resemblance. Indirect alignment MA-NCI using FMA as a reference ontology The following method was used for automatically deriving a mapping between
MA and NCI from the two direct alignments MA-FMA and NCI-FMA. When
a FMA concept CF is aligned with both a MA concept ({MA: CM, FMA: CF}) and a NCI concept ({NCI: CN, FMA: CF}), the concepts CM and CN are automatically aligned ({MA: CM, NCI: CN}). For example, as shown in , the direct alignment MA-FMA identifies the match {MA: Forelimb, FMA: Upper limb (synonym: Forelimb)}, which is supported by positive evidence. The direct alignment
NCI-FMA identifies the match {NCI: Upper extremity, FMA: Upper limb (synonym: Upper extremity)}, also supported by positive evidence Therefore, the match {MA: Forelimb, NCI: Upper extremity} is derived automatically, through the FMA concept Upper limb, supported by positive structural evidence in both direct alignments. Comparing two alignments between MA and NCI We compared the matches obtained by direct alignment MA-NCI and by indirect
alignment through the FMA. The matches were classified into three
groups: matches identified by both alignments; matches specific to the
direct alignment MA-NCI; and matches specific to the alignment through
the FMA. As shown in , the match {MA: Forelimb, NCI: Upper extremity} belongs to the first group. Further analysis of structural evidence
for the matches was performed in the three groups. Three direct alignments Results for three direct alignments are summarized in Table 1. The alignment NCI-FMA yielded the largest number of matches (2,173) and
MA-NCI the smallest (715). A very small number of conflicts was identified
in the two direct alignments to FMA; none in the direct MA-NCI
alignment. In the three direct alignments, a vast majority of the matches (> 90%) was supported by positive structural evidence. No
evidence (positive or negative) was found for 5–9% of
the matches in three direct alignments. For example, although Elbow joint has relations to other matches in both MA (e.g., PART-OF Forelimb) and NCI (e.g., PART-OF Skeletal system), none of these relations are shared. Indirect alignment MA-NCI through FMA 703 matches between MA and NCI were automatically derived from the 1,353 matches
in the direct alignment MA-FMA and the 2,173 matches in NCI-FMA. 649 of
them (92%) received positive structural evidence in
both direct alignments MA-FMA and NCI-FMA, 8 (1%) received negative
evidence in one of the two direct alignments, and 46 (7%) received
no evidence in at least one of the two direct alignments. Comparison between direct and indirect alignment for MA-NCI Quantitative results We compared the 715 matches obtained in the direct alignment MA-NCI to
the 703 matches resulting from the indirect alignment through the FMA. The
results of this comparison are summarized in . The most important finding is that 654 matches are shared by both alignments, leaving 61 matches specific to the direct alignment and 49 specific
to the indirect alignment through the FMA. Matches supported by structural evidence Among the 654 shared matches, 583 (89%) received positive structural
evidence in all three direct alignments, e.g., {MA: Forelimb, NCI: Upper extremity}. Matches without structural evidence 65 (10%) of the 654 shared matches received no structural evidence
in at least one of the three direct alignments. Although linked to
other matches in MA (e.g., PART-OF Cranium), Chondrocranium has no hierarchical relations to any other matches in NCI and FMA. This
is why the matches of Chondrocranium receive no evidence in any of the three direct alignments. Conflicts At last, 6 (1%) of the 654 shared matches received negative evidence
in one of the three direct alignments. For example, while a concept Pericardial cavity is present in the three ontologies, the corresponding match received negative
evidence in the direct MA-FMA alignment, no evidence in MA-NCI, and
positive evidence in NCI-FMA. Because of the presence of a conflict
in MA-FMA, no match is identified in the indirect alignment through
FMA. In this case, the indirect alignment MA-NCI (no match suggested) is
consistent with the direct alignment MA-NCI (match not supported
by structural evidence). In fact, domain knowledge is required to evaluate
the match in these cases. Why are some matches in the alignment through the FMA not identified in
the direct alignment? 49 matches in the indirect alignment through the FMA (7%) are not
identified in the direct alignment between MA and NCI, most of them
being valid (supported by structural evidence). The analysis of these
cases reveals two major causes. Additional synonyms First, the FMA records a much larger number of synonyms for anatomical
entities than the other two ontologies, increasing the chances of finding
a lexical match between names in these ontologies and the FMA. For
example, the terms Integumental system in MA and Integumentary system in NCI do not match directly, but are listed as synonyms in the FMA. Additional relations The FMA also provides a much larger number of relations among anatomical
entities, amplified by the inference techniques used in our method. The
presence of these additional relations increases the chances of finding
similar relations between an ontology and the FMA and maximizes
the chances for matches to receive positive structural evidence. It also
increases the chances of identifying conflicts, therefore enhancing
the overall quality of the mappings. Why are some matches in the direct alignment not identified in the alignment
through the FMA? 61 matches in the direct alignment between MA and NCI (nearly 9%) are
not identified in the indirect alignment through the FMA, most
of them being valid (supported by structural evidence). Analogously, among
the 44 matches identified both directly and indirectly, but not supported
by structural evidence in the indirect alignment, 14 received
positive evidence in the direct alignment. Differing coverage and representation
of anatomy seem to be the cause of these differences. Differing coverage Some concepts present in MA and NCI are absent from the FMA. This is the
case, for example, of Iliac artery. While Common iliac artery, Internal iliac artery and External iliac artery are present in all three ontologies, Iliac artery is present in MA and NCI, but absent from the FMA. Thus, this concept
can be aligned directly, but not through the FMA. We leave anatomists
to decide whether and how iliac arteries should be represented. Such discrepancies
in coverage are indicative of some potential problem and
should be reviewed. Differing representation Unlike particular veins and arteries, general concepts such as Blood vessel are defined as a General anatomical term in the FMA and do not form the root of a hierarchy of blood vessels, as
it is the case in MA and NCI. Therefore, the relations of particular
veins and arteries to Blood vessel present in MA and NCI are not shared with FMA, although the concepts themselves
are present. Again, automatic alignment techniques can at best
identify such issues. Normalizing the representation generally requires
domain experts. Alignment through a reference ontology vs. pairwise alignment As suggested in the introduction, mapping through a reference ontology
is cost-effective: n ontologies require n(n-1)/2 paiwise mappings, but only n-1 mappings to a reference ontology. As illustrated in , for five ontologies – which is a small number by Semantic Web
standards – the difference already represents a 60% economy (4 vs. 10). This study confirms the feasibility and efficiency of indirect alignment
through a reference ontology. Of the 715 matches identified by direct
alignment, 654 (91%) have been discovered by the indirect alignment. Moreover, the
indirect alignment was able to identify matches
not discovered by direct alignment. Overall, this study suggests that
the performance of the indirect alignments is consistent with – if
not better than – that of direct alignments. The indirect alignment assumes the existence of an ontology that can serve
as reference. Desirable characteristics for such an ontology include
broad coverage (in terms of both concepts and relations), inclusion
of many synonyms and compatibility with standard representation principles. In
our experiment, we found the FMA to have many of these characteristics: its
large size and comprehensive set of synonyms certainly
contributed to the high percentage of mappings discovered (compared to
direct alignment) and outweigh its idiosyncrasies. Current limitations and future work The MA-NCI alignments, direct and indirect, have identified a total of 764 matches. These
only account for about one third of the concepts in
MA and NCI anatomical concepts (excluding the 2000 NCI concepts for cell
types and subcellular components, not represented in MA). Our alignment
approach relies heavily on the lexical similarity and is limited
to the identification of one-to-one concept matches. We plan to investigate
complementary approaches based on structural similarity, as well
as complex matches (one-to-many and many-to-many). The three ontologies aligned in this study all represent the same domain: vertebrate
anatomy. The analysis of fine differences between human
and mouse anatomies is beyond the scope of this paper but is addressed
in [ 8]. The absence of validation of the alignment obtained by our fully automatic
techniques is another limitation of this study. The manual review
of the matches by an expert is labor intensive and costly. While manual
validation remains an objective of this project, we believe that the
comparison between direct and indirect alignments provides some elements
of cross-validation of the results. This study suggests that, when a reference ontology exists, indirect alignment
of multiple ontologies through a reference represents a valid, cost-effective
alternative to pairwise alignment. We believe that reference
ontologies will be a key component of semantic integration of biomedical
information and interoperability of biomedical applications. Besides
anatomy, biochemistry is one of the domains which would benefit
most from the development of reference ontologies (e.g., an ontology
of small molecules). The research was supported in part by an appointment to the National Library
of Medicine Research Participation Program administrated by the
Oak Ridge Institute of Science and Education through an interagency agreement
between the U.S. Department of Energy and the National Library
of Medicine. Thanks for their support to Cornelius Rosse and José Mejino
for the Foundational Model of Anatomy, Terry Hayamizu for
the Adult Mouse Anatomical Dictionary, and Sherri De Coronado for the
NCI Thesaurus. 2. Rector AL, Bechhofer S, Goble CA, Horrocks I, Now-lan WA, Solomon WD. The GRAIL concept modelling language for medical terminology. Artif Intell Med. 1997;9(2):139–71. [PubMed] 3. Noy NF. Tools for mapping and merging ontologies. In: Staab S, Studer R, editors. Handbook
on Ontologies: Springer-Verlag; 2004. p. 365–384. 4. Hayamizu T, Mangan M, Corradi J, Kadin J, Ringwald M. The Adult Mouse Anatomical Dictionary: a tool for annotating and integrating
data. Genome Biology. 2005;6(3):R29. [PubMed] 5. De Coronado S, Haber MW, Sioutos N, Tuttle MS, Wright LW. NCI Thesaurus: Using Science-based Terminology to Integrate Cancer Research
Results. Medinfo. 2004;2004:33–7. [PubMed] 6. Rosse C, Mejino JL., Jr A reference ontology for biomedical informatics: the Foundational Model
of Anatomy. J Biomed Inform. 2003;36(6):478–500. [PubMed] 7. Zhang S, Bodenreider O. Aligning representations of anatomy using lexical
and structural methods. AMIA Annu Symp Proc 2003:753–7. 8. Travillian RS, Gennari JH, Shapiro LG. Of mice and men: Design a comparative
anatomy information system. AMIA Annu Symp Proc 2005:(in press).
|
This article has been 
