Bioinformatics Advance Access originally published online on March 29, 2005
Bioinformatics 2005 21(11):2773-2779; doi:10.1093/bioinformatics/bti409
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Formalizing concepts of species, sex and developmental stage in anatomical ontologies
School of Informatics, The University of Edinburgh Edinburgh EH8 9LE, United Kingdom
 |
Abstract |
|---|
 TOP Abstract INTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
Motivation: Anatomy ontologies have a growing role in bioinformatics—for example, in indexing gene expression data in model organisms. To relate or draw conclusions from data so indexed, anatomy ontologies must be equipped with the formal vocabulary that would allow statements about meronomy to be qualified by constraints such as part of the male or part at the embryonic stage. Lacking such a vocabulary, anatomists have built this information into the structure of the ontology or into anatomical terms. For example, in the FlyBase anatomy for drosophila, the term larval abdominal segment encodes the stage in the term, while the terms male genital disc and female genital disc encode the sex. It remains implicit that a fly has one and only one of these parts during its larval stage. Such indicators of context can and should be represented explicitly in the ontology.
Results: The framework we have defined for anatomical ontologies allows the canonical anatomy structures of a given species to be those common to all sexes, and to have either male, female or hermaphrodite parts—but not combinations of the latter. Temporal aspects of development are addressed by associating a stage with organism parts and requiring a connected anatomy to have parts that exist at a common stage. Both sex and anatomical stage are represented by attributes. This formalization clarifies ontological structure and meaning and increases the capacity for formal reasoning about anatomy. The framework also supports generalizations such as vertebrate and invertebrate, thereby allowing the representation of anatomical structures that are common across a sub-phylum.
Availability: http://www.aiai.ed.ac.uk/resources/bioinf/
Contact: stuart{at}aiai.ed.ac.uk
Supplementary information: http://www.aiai.ed.ac.uk/resources/bioinf/
|
INTRODUCTION |
|---|
|
TOPAbstractINTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
The Gene Ontology (GO) provides a species-independent vocabulary for annotating the results of gene expression assays (GO Consortium, 2000). Where such assays are tissue-specific, they can also be annotated with or indexed by terms from anatomical ontologies for the model species (Bard and Rhee, 2004). These anatomies describe anatomical structures and tissue types, and are available under the Open Biological Ontologies (OBO) initiative (http://obo.source-forge.net).
Describing and reasoning about anatomy are complex tasks which, to-date, have been approached from a largely intuitive, pragmatic perspective for the model species. Differences in anatomical modelling can be resolved by clarifying the definition and use of the is_a and part-of relations in anatomy ontologies (Schulz and Hahn, 2004; Smith et al., 2003). The resulting benefits include enhanced sharing of ontologies, increased possibilities to map between ontologies and the potential to automate reasoning about anatomy. The resulting improvement in computational interoperability will increase the uses to which ontologically annotated data can be put (Gennari et al., 2005).
This paper addresses three other aspects of the largely intuitive modelling of anatomy, namely, modelling species, sex and developmental stage. These have been realized in an ad hoc, inconsistent or implicit manner—unsuitable for automated reasoning. For example, in an OBO anatomy, species is typically denoted by the root node in the ontology. Below this node, the parts and classes of terms are species-specific: nervous system in the drosophila anatomy ontology is to be interpreted as a particular subset of organ system that is part-of a drosophila. This approach works if one is modelling the physical parts of an individual species. However, it fails to account for developmental processes and events, as these cannot be modelled as part-of the organism, or types of an organism. An integrated framework for anatomy must properly account for these entities rather than blur the meaning of is-a and part-of. Combining anatomy ontologies in the simple manner of combining the (typically) species-identifying top nodes of anatomy ontologies into a multi-species part-of graph is not an adequate means of putting them into a common framework. For example, the non-physical anatomical entities of the foundational model of anatomy (FMA) (Rosse et al., 1998; Rosse and Mejino, 2003), which include the part-of relations and other representational vocabulary, are not considered part-of the human (and they are not species-specific) and so could not be integrated with other ontologies in such a fashion.
With respect to sex and developmental stage, within an OBO anatomy for a single species, these have usually been indicated by adjectives. But this is not suitable for computer-based reasoning. For example, the terms embryonic and male are used to qualify the timing or sex-specificity of drosophila tissues. Timing can also be represented by stating that the entity is part-of an ‘organism at a stage’, while an explicit descends relation is used in mouse and drosophila anatomies to indicate some developmental orderings between tissues. However, the primary means to give timing information about the tissues being so related is by introducing an appropriate adjective.
In combination, these practices are a barrier to reasoning about anatomy, and obscure the commonalities that exist across species, e.g. that digestive and reproductive systems can be identified in drosophila, Caenorhabditis elegans, mouse and human (albeit in different instantiations). Following the distinction in Rosse and Mejino (2003) between canonical anatomy, which is concerned with the synthesis of generalizations of observed anatomical structures, and instantiated anatomy, which is concerned with the anatomical data of individual organisms, for example, the right atrium in my body, we have developed and implemented a formal solution to the problem of representing sex and stage in canonical anatomy.
This is not an exercise in formality for formality's sake: rather, it supports, for example, the automated extraction of developmental patterns from data annotated with terms from anatomical ontologies as well as terms from GO, such as the molecular pathway for sex determination in drosophila. In a fly larva, the production of a functional Tra protein together with the expression of tra2 represses male differentiation genes and leads to female development. This occurs in the male repressed primordium, part of the female genital disc. This gene is not expressed in the male but is listed in FlyBase as being present in the male genital disc, carrying the free text annotation ‘female’. The meaning of the annotation is clear to the human reader, but an automated analysis would be unable to identify male- or female-specific expression patterns without explicit representation of the important attributes.
In ontological modelling, the distinction between is_a and part-of is fundamental to modelling anatomy. Because our work builds on an established theory of parthood, which we incorporate into an ontology of concepts, we complete this introduction with a discussion of the is_a and part-of relations. Henceforth, we will use ‘class’, ‘category’ and ‘type’ as synonyms for concept, a sans-serif font for specific concepts, e.g. 
, and italics for relationships, e.g. partOf. Where we are discussing the part–whole relation in general terms, we will use ‘part-of’. More specific senses will be denoted by a relationship given in italics.
The meaning of the is_a relationship between concepts in an ontology can be thought of in terms of sets. Figure 1 illustrates the view of the relationship between 
and its subclasses (or subsets) , 
and 
as defined in the foundational model of anatomy (Rosse et al., 1998). It can be seen that all instances of are also instances of (where a dot represents a concrete instance).
|
In modelling anatomy, the critical issue is how the sets of parts and wholes are related. Specifically, the question is what constraints are placed on the part-of relation when we state that
is part-of ? We might require that all instances of have a , or that all instances of are part-of a , or both. Note that part-of must be defined in terms of instances, and that this relation is quite different from subset.
Two separate formalizations of the part–whole relation are illustrated in Figure 1 where the block arrows indicate that can be defined as a part-for (Aitken et al., 2004; Smith and Rosse, 2004), or that can be defined in terms of the concept by the has-part relation (Aitken et al., 2004; Smith and Rosse, 2004). Note that these are class-level relations.
The part-for constraint is also shown in Figure 2a where all instances of are elements of a set of all heart parts. Similarly, all instances of 
are members of this set. In this case, the parts must be part-of some whole, but not necessarily the same whole. Under the part-for constraint, anatomical parts are defined in terms of the wholes they belong to.
|
Figure 2b shows the has-part constraint where all instances of
have some instance of each part-class as a part. The has-part constraint specifies the canonical anatomy tree that we would expect, where each instance of has a and a as parts. The whole is (partially) defined in terms of its parts. Parts of a given type may be parts of several different types of wholes, as the definition describes the whole rather than the part.
The two separate constraints have implications for the way in which anatomical parts are named and modelled. part-for entails that entities are defined by their position in the anatomy. For parts such as and , this is appropriate. On the other hand, it is inappropriate for many anatomical parts—in particular, mesenchymal tissue, skin and blood that are distributed throughout the organism. For these, only has-part captures the desired relationship. has-part can be used to relate sex-specific parts such as mouse testis to the male mouse, but not to the class of all mice.
In our formal solution to specifying species, sex and developmental stage in anatomy ontologies, we specify the structure of instance-level part-of relations as connected graphs, as illustrated in Figure 2b for a simple case. To constrain the anatomy to include such canonical structures, we adopt the has-part constraint as the primary interpretation of part-of. The additional part-for interpretation can also be stated where applicable.1 The anatomy is said to be a specification as it is a generic description holding for all prototypical instances, rather than the description of a specific individual.
We now describe the Anatomy Context Ontology (ACO) which is a top-level ontology that allows the distinctions of species, sex and stage to be incorporated into anatomy. This ontology is grounded in an established formal theory of parts, which we also present. Next we address the engineering issues of using the Web Ontology Language (OWL) as a means of expressing and exchanging the ontology. Use of OWL permits the reuse of many tools and modelling techniques. Finally, the formal verification of the ontological definitions and issues of inferencing are described.
|
MATERIALS |
|---|
|
TOPAbstractINTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
The anatomy ontologies analysed in this paper are taken from publicly available sources. The revised ontologies are provided as a community resource on the supplementary materials website. The specific anatomies we consider here are: drosophila (Ashburner) and C.elegans (Lee), both obtained from OBO. The developmental mouse (Kaufmann, Bard and Davidson) and developmental human (Bard) anatomies are derived from part-of and stage data held in the MRC Mouse Atlas database http://genex.hgu.mrc.ac.uk/. These were provided by Jonathan Bard and Albert Burger. The revisions were performed by the author in consultation with Bard. (The supplementary materials can be found at: http://www.aiai.ed.ac.uk/resources/bioinf/.)
|
THE ANATOMY CONTEXT ONTOLOGY: ONTOLOGICAL MODELLING AND ENGINEERING METHODOLOGY |
|---|
|
TOPAbstractINTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
Formalizing concepts of species and sex
In the ACO, we propose a relatively small vocabulary of concepts and definitions that capture generic anatomical notions such as species, sex and developmental stage. These proposals do not aim to address the principles for organizing hierarchies of anatomical concepts (but see Rogers and Rector, 2000; Rosse and Mejino, 2003; Noy et al., 2004; Kumar et al., 2005). However, there are implications for categorization that follow from the view of part-of that is adopted, as part-for can be taken to be definitional, as discussed earlier.
A rich ontology such as the FMA has an extensive categorization of anatomical entities and a set of criteria for modelling the domain. In contrast, anatomical ontologies for the model species either lack a connected is_a hierarchy (e.g. drosophila and C.elegans) or lack any is_a relation (e.g. mouse). Therefore, it is necessary to introduce high-level concepts such as 
or 
in order to provide an appropriate category for these classes of entities. The class-level relations can then be properly defined, for example, to allow part-of relations between and , but to exclude part-of relations between and 
. It can be seen that we are not proposing a new anatomy: rather we are attempting to clarify the formal issues that arise in modelling anatomy symbolically. In the following presentation, we use subClassOf for the concept subsumption relationship is_a, as it has a formal definition in OWL that we make use of later.
We start with the formal issues involved in modelling species. Here we use the conventional notions of species and subspecies: species can be organized into genus, family, order, class and phylum, and ultimately into kingdom and super kingdom. For example, the species Mus musculus (house mouse) belongs to the genus Mus, family Muridae, order Rodentia and class Mammalia (Parker, 1982). 2 The southeastern Asian house mouse Mus musculus castaneus is a subspecies of Mus musculus—see Lundrigan et al. (2002) for an analysis of mouse phylogeny based on DNA sequences. We do not pursue the taxonomic organization of species further in this paper. However, this taxonomy will be useful for organizing anatomical properties and structures at genus, order or phylum level. For our purposes, species are subclasses of 
being entities of a specific genotype.
Before discussing the formal issues involved in modelling aspects of anatomy related to the sex of an organism, we briefly review the relevant biology. Many organisms are sexually dimorphic, giving rise to males and females of the species that produce sperm and eggs (gametes) (Gilbert, 2000). Others are hermaphrodite, where both gametes can be produced by a single individual. C.elegans is also dimorphic, but is unusual in giving rise to hermaphrodites and males. The primary sex characteristic of an organism is that of the gonads (testis or ovary) (Müller, 1996). Secondary sex characteristics, including size and markings, may be governed hormonally or genetically. In drosophila and C.elegans, sex is determined by the ratio of the X chromosomes to autosomes, rather than by the presence of a Y chromosome. In drosophila, the sex of somatic cells—and therefore the secondary sex characteristics—is determined autonomously at the cellular level. In contrast, mammalian cells outside the gonads do not have male or female characteristics. If we consider the potential use of the ontology in clinical information systems such as those dealing with patient records, we would encounter transgender issues. However, we exclude these characteristics at present.
To account for the complexity of the biology in an anatomical ontology, we assign the characteristic of being male, female or hermaphrodite to the whole organism, rather than to cells and tissues. In certain species, sex characteristics can be assigned to cells, and can be manipulated experimentally, but they need not be accounted for at the cellular or tissue level in prototypical anatomy. Following the FMA, the category is the most general class of anatomical parts, including organs and organ systems, as well as body substances and anatomical spaces. We define it to be disjoint from the category of whole organisms (a distinction not made in the FMA). Figure 3 shows the top-level concepts and the categorization of organisms by species. The category 
is partitioned according to sex, and this partition is exhaustive. All mice must be male or female, but not both. This partition is inferred from the attribute sex that relates instances of to the attribute values that denote the primary sex characteristics of the whole organism. While such attributes are not shown in Figure 3, they are involved in axioms that further elaborate the ontology.
|
For example, axioms (1) and (2) state that the subcategories of the class
are partitioned by means of attribute values 
and 
. Since is a subclass of (3) it then follows by inference that is similarly divided into male and female. Named classes are introduced to represent these partitions (4,5):3
 | (1) |
 | (2) |
 | (3) |
 | (4) |
 | (5) |
We define all anatomical entities to be part-of a single organism instance. That is, the anatomy tree is rooted at an instance of type . The theory of parts and wholes is put on a more formal basis in the following section, where we motivate the formalization by reference to the underlying biology, existing anatomical modelling practice and the way in which embryonic development is studied.
Formalizing concepts of sex-specific parts
We need to capture not just the sex of an organism instance, but also its sex-specific parts. Axioms (6–11) are a standard axiomatization of the part-of relation P. This relation is reflexive (6), transitive (7) and antisymmetric (8). The proper part relation PP adds the property of irreflexivity to part-of (9). The relationship symbols in this theory are abbreviated as below:
Direct parts DP and overlap O are defined in terms of PP and P [axioms (10) and (11)].
 | (6) |
 | (7) |
 | (8) |
 | (9) |
 | (10) |
 | (11) |
These axioms are defined in terms of properties that hold between instances. However, if we wish to represent the part-of relationship between concepts (i.e. classes or types) we need to define appropriate class-level relationships. As the OBO anatomies for the model species already use a part-of relation that holds between concepts, but do not define what this relation means in terms of instances, it is worthwhile to formalize the interpretation. This is further discussed in Aitken et al. (2004) where the binary relations in Table 1 are defined.
|
To illustrate the approach, axiom (12) defines partOf when qualified by the classDefinition relation, stating that all instances of the whole-class (denoted by the type relation) have some instance of the part-class as a proper part. This corresponds to the has-part constraint introduced earlier.
 | (12) |
The relations in Table 1 were derived from an analysis of existing OBO anatomies. They are a means to formalize current practice in anatomical modelling. They allow proper formalization where all definitions would be classDefinitions with part-of inherited to subcategories, but also allow for the case where this inference is not intended. A fuller attempt to capture the underlying biological structures might add a surface-of relation and distinguish other part-of and connected-to relationships. We do not address these questions in this paper, but note that during development, anatomical structures do not have the fully formed cavities and connected structures found in the adult, and hence spatial form and inclusion may be less useful as an organizing principle. Further senses of the part–whole relation are the component–integral sense where the part bears a functional relation to the whole, and the member–collection sense where the part is a member of a group. For example, the vertebrae can be considered to form a group. A total of six senses are identified by Winston et al. (1987), who note that when the senses are mixed, the property of transitivity is lost. We assume that part-of is used in a single sense in the ontologies we consider. Should the ontology be extended with additional meronymic relations, then each of these will have the properties (6)–(8) defined above, but the subsuming parts relation will not be transitive.
Consider now the part–whole relation for male and female organisms—defining to be disjoint with means statements about part-of can be made that relate a part to the whole entity that it is a component of, and to the organism type it is part of without ambiguity. For example, 
is part-of the 
(entity) and part-of the 
(organism). The axiom that interprets the part-of statements must be revised to account for part-of statements. Axiom (12) is replaced by axiom (13):
 | (13) |
This formula places an extra condition on the wholes w that have proper part p—they must also be part of an organism s of a type S. When this extra condition is satisfied the whole w has p as a proper part. Figure 4 shows the partOf relationships for 
which is part-of all mice, and which is only part-of the male mouse. The anatomy tree for the male mouse has all parts of and the male-specific parts, but no female-specific parts. The full listings of the mouse and human anatomy ontologies represented in the manner of Figure 4, containing 3525 and 2327 terms respectively, are provided in the supplementary material.
|
Formalizing the concept of stage
The development of model organisms is studied by organizing the time period from fertilization to birth into developmental stages. In drosophila, the major stages correspond to major morphological steps: from egg through pupa to adult. In mouse, the staging is defined according to the 28 Theiler stages (Kaufmann and Bard, 1999), and in human the 23 Carnegie stages are used. In zebrafish, seven periods (zygote, cleavage, blastula, gastrula, segmentation, pharyngula and hatching) are further divided into stages (Kimmel et al., 1995). The embryonic organism develops continuously, but embryonic development is studied, and the results archived, on the basis of one of the prototypical staging schemes applicable to the organism in question.
Formally, we model developmental stage as an attribute value. In this case the attributes hold of and terms representing the stages over which the entity is part-of the whole. The class-level attributes startStage and endStage relate subclasses of to 
denoting the first and last stage, respectively, where that part occurs. The attribute stage relates to each stage where it occurs, derived from its start and end stages. As above, the interpretation of part-of is modified to include the condition that the part entity and whole entity have the same stage attribute; i.e. that they exist at the same stage. Stages are canonical in the same sense that the anatomical structures are: we can consider a stage as an abstraction of a set of temporal intervals that are prototypical of a developing organism. The start and end of an interval is denoted by a time point for a particular instance of an organism; thus there is a mapping to the real time line. Figure 4 illustrates stage information associated with (a) the heart and one of its parts, and (b) the male reproductive system of the mouse and one of its parts.
The part-of relation is given a temporal interpretation through ordering stage attributes such as Theiler Stages by the relation <s. An ordering over temporal intervals,4 <i, is defined in terms of <s. (Both ordering relations are irreflexive and transitive.) We define a function 
as returning a temporal interval given an instance of and a stage attribute. The function represents the temporal interval during which an organism instance passes through a stage. The relationship between , <i and <s is defined as:
 | (14) |
 | (15) |
The part–whole relations are indexed by the time interval over which they hold, and the ordering of these intervals is determined by the prototypical staging of the organism-type in question. Full definitions of the relationships and functions needed to formalize these concepts are given in the supplementary material.
A class–level descends relation can be introduced to represent developmental lineage; this relation is defined in terms of the stage attributes: The start stage of the earlier developing tissue must be <s the start stage of the later developing tissue.
The task of authoring axioms is not one that most biologists would expect to perform, but this is not required in the proposed ontology. The axioms are schema that define the meaning of concrete assertions, e.g. the assertions shown diagrammatically in Figure 4. That is, the knowledge engineer, or anatomist, can author knowledge solely in terms of the class-level vocabulary of partOf, startStage and endStage.
The Anatomy Context Ontology in OWL
The ACO that has been presented here is defined in a first-order logic, KIF (Genesereth and Fikes, 1992). The intended meaning of classes and relationships is represented with a minimal bias from the encoding language. However, KIF is not currently the prevalent choice for ontology exchange or tool development. Therefore, we specify an OWL encoding of the ontology to address these issues. Because the class-level vocabulary is represented by binary relations, the OWL representation can be constructed directly from the classes and relations introduced earlier.
While the ACO ontology might have made a greater use of ternary relationships or functional terms, we minimized their use (eliminating it from the class-level vocabulary) in order to make it easier to express in OWL. Based on RDFS, OWL represents relations as binary relations (usually called predicates) between the subject and the object of the relation. The simplest unit is the triple: <Subject Predicate Object>. A collection of triples forms a graph which can be interpreted as an RDFS or OWL ontology. It is clear that there is a syntactic difference between RDFS and KIF. In addition, RDFS treats relations as objects, and so this must be accounted for when RDFS graphs are interpreted.
Relations and classes can be defined in OWL syntax, and concrete facts can be stated. For example, facts about are expressed by triples whose subject is the identifier for this term (EMAPA:16106) and the identifiers for the heart (EMAPA:1610) and the start and end stages (XUO:0000055 and XUO:0000056).
<owl:Class rdf:ID="EMAPA.16106"><name>cardiogenic plate</name>
<!-- is partOf the Heart EMAPA.1610 -->
<partOf rdf:resource="#EMAPA.1610"/>
<!-- from TheilerStage 11 to 12 -->
<startStage rdf:resource="#XUO.0000055"/>
<endStage rdf:resource="#XUO.0000056"/>
</owl:Class>
The XSPAN project (http://www.xspan.org) is already using this OWL ontology for data exchange between tools.
The axioms of the ontology cannot easily be stated in RDFS or OWL. Neither can the definitions be fully expressed in the description logic fragment of OWL (OWL-DL). Therefore we follow the RDFS approach to interpreting the RDF graph: the class-level relations are represented in OWL as triples, and interpreted by the axioms of the theory using the translation from RDFS to first-order logic described in McGuinness et al. (2002). The translation rules are as follows:
 | (16) |
 | (17) |
|
VERIFICATION AND INFERENCE |
|---|
|
TOPAbstractINTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
Verification ensures that the axioms are correct and consistent, and also allows one to prove interesting properties of the class-level relations. We address the latter in this section. The interesting properties to show are that partOf is transitive and that other assertions of partOf can be inferred. Once these properties are established, we can limit the reasoning task to reasoning with this subset of the ontological relations.
Using the Otter theorem prover (Wos, 1996) we establish the transitivity of partOf:
 | (18) |
Having proven axiom (18), we can apply it to the OWL representation of the ontology in order to compute the transitive inferences that follow from the concrete partOf statements that define the anatomy (i.e. to find all sub-parts). In Java applications, such rules can be implemented using the inference methods of the Jena Semantic Web toolkit (http://www.hpl.hp.com/semweb/jena.htm). Rule (18) can be applied to an entire ontology in the order of seconds, to hundreds of seconds, depending on the size of the ontology. Therefore, we have a solution to the engineering problems of ontology exchange and tool building, and a practical means to reason about the ontology.
The COBrA ontology editor (Aitken et al., 2005; Korf, 2003) provides an interface to the ruled-based reasoner in the Jena toolkit, and is able to read and write OWL. Thus COBrA can be used to create and edit ontologies, and also check constraints and answer queries through making inferences.
|
RELATED WORK |
|---|
|
TOPAbstractINTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
Returning to the issue of the conceptualization, in the ACO, developmental stage is not used to partition the subspecies category. If we were to represent the developmental mouse in this manner, there would be 28 categories of mouse, further divided into male and female after Theiler stage 21. In drosophila, the partition of the organism into six stages (embryo, egg, larva, prepupa, pupa and adult) is a more natural modelling decision. However, there is no information associated with the 20 sub-stages of embryo, or with other sub-stages. To provide these anatomical models would require duplicating the anatomy tree in every sub-stage, thus creating many hundreds of identical concepts for structures such as organ systems, and for each substructure.
It is likely (but not explicitly stated) that the six categories of the drosophila organism are to be considered to be mutually exclusive. If so, no part can be shared across two successive stages (at the same time), as then the anatomy tree would have multiple root nodes and this is usually taken to be undesirable. This leads to the problem of duplicating the anatomy tree. If not, then we have the situation that exists in the abstract mouse (Burger et al., 2004) where the organism has all of its parts at all stages simultaneously.
The assumption may be that the class membership relation type is indexed by time, and so the same entity can belong to different categories at different time points. This approach has been advocated by Smith and Rosse (2004). However, such a step has implications for the other relationships in the ontology, as they may also require a temporal index. This adds significant complexity to systems engineering, as established languages do not allow the type relation to take a temporal argument. We also object to this approach on the conceptual grounds that the stages are canonical. In modelling development, it is more appropriate to use stage as an intermediate ‘index’ for the interval over which the part–whole relation exists, than to use ontological categories for ‘organism at a stage’ and rely on an index from entities to the real time-line directly.
On the issue of a methodology for formal ontology development, we note that proposals for sets of ontological axioms often fail to address the engineering and tool-building problems, and do not have a practical solution to making inferences based on the axioms (Bittner, 2004; Bittner and Donnelly, 2004). Relatively efficient reasoners exist for description logics and these have also been used to represent anatomy (Hahn et al., 1999; Rogers and Rector, 2000). However, due to reduced expressivity, these logics impose a bias for the part-for view of anatomy structure, or require the introduction of additional terms to distinguish the whole entity and the set of its parts. The frame-based approach (Noy et al., 2004) is able to address both formal and engineering aspects.
|
CONCLUSION |
|---|
|
TOPAbstractINTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
The proposed ACO makes explicit the concepts of species, sex and developmental stage. These notions are typically built into the names of anatomical entities, where they remain inaccessible to automated reasoning. In addition to providing a remedy to this problem, the proposed ontology puts the anatomies of individual species into a wider context. Extending the approach, we can incorporate generalizations such as vertebrate and invertebrate into the
hierarchy and define shared anatomical structures at this sub-phylum level.
|
|
Acknowledgments |
|---|
The author is grateful to Jonathan Bard, Bonnie Webber and Albert Burger for their help with the preparation of this paper, and for providing the source data. This work is supported by BBSRC grant BBSRC 15/BEP 17046 (XSPAN), and under the Advanced Knowledge Technologies (AKT) Interdisciplinary Research Collaboration (IRC), which is sponsored by the UK Engineering and Physical Sciences Research Council under grant number GR/N15764/01. The AKT IRC comprises the Universities of Aberdeen, Edinburgh, Sheffield, Southampton and the Open University.
|
Footnotes |
|---|
1In Smith and Rosse, 2004 the part-of relation holds when both part-for and has-part hold for two classes.
2Terms from the NCBI Taxonomy http://www.ncbi.nlm.nih.gov/Taxonomy/ can be incorporated where they correspond to primary sources.
3Technical note: Attribute values are sets of instances. They may be totally or partially ordered, and may assign a unique value to an object. The relation type holds between an instance and the category it belongs to.
4A temporal interval is defined by its start and end time point, and intervals are ordered by the relative order of their start/end times.
Received on December 17, 2004; revised on January 3, 2005; accepted on March 22, 2005
|
REFERENCES |
|---|
|
TOPAbstractINTRODUCTIONMATERIALSTHE ANATOMY CONTEXT ONTOLOGY:...VERIFICATION AND INFERENCERELATED WORKCONCLUSIONREFERENCES |
|---|
Aitken, J.S., et al. (2004) Part-of relations in anatomy ontologies: a proposal for RDFS and OWL formalisations. Proc. PSB, 9, 166–177.
Aitken, S., et al. (2005) COBrA: a bio-ontology editor. Bioinformatics, 21, 825–826
Bard, J.B.L. and Rhee, S.Y. (2004) Ontologies in biology: design, applications and future challenges. Nat. Rev. Genet., 5, 213–222[CrossRef][ISI][Medline].
Bittner, T. (2004) Axioms for parthood and containment relations in bio-ontologies. Proc. KR-Med., 4–11.
Bittner, T. and Donnelly, M. (2004) Proceedings of the European Conference on Artificial Intelligence. The mereology of stages and persistent entities. Proc. ECAI, 283–287.
Burger, A., et al. (2004) Formalization of mouse embyro anatomy. Bioinformatics, 20, 259–267
Genesereth, M.R. and Fikes, R.E. (1992) Knowledge Interchange Format, Version 3.0, Reference Manual. Technical Report Logic-92–1, , Stanford Computer Science Department, Stanford University.
Gennari, J.H., et al. (2005) Integrating genomic knowledge sources through an anatomy ontology. Proc. PSB, 10, 115–126.
GO Consortium. (2000) Gene Ontology: tool for the unification of biology. Nat. Genet., 25, 25–29[CrossRef][ISI][Medline].
Gilbert, S.F. (2000) Developmental Biology. Sinaur Associates.
Hahn, U., et al. (1999) Part-whole reasoning: a case study in medical ontology engineering. IEEE Intell. Syst., 14, 59–67.
Kaufmann, M.H. and Bard, J.B.L. (1999) The Anatomical Basis of Mouse Development. Academic Press.
Kimmel, C.B., et al. (1995) Stages of embryonic development of the zebrafish. Dev. Dyn., 203, 253–310[ISI][Medline].
Korf, R. (2003) COBrA—A Concept Ontology Browser for Anatomy. Informatics Report EDI-INF-IM030022, University of Edinburgh.
Kumar, A., et al. (2005) Biomedical informatics and granularity. Comparat. Func. Genom., 5, 501–508.
Lundrigan, B.L., et al. (2002) Phylogenetic relationships in the genus Mus, based on paternally, maternally, and biparentally inherited characters. Syst. Biol., 51, 410–431[CrossRef][ISI][Medline].
McGuinness, D.L., et al. (2002) DAML+OIL: an ontology language for the semantic web. IEEE Intell. Syst., 17, 72–80.
Müller, W.A. Developmental Biology, (1996) , Berlin Springer.
Noy, N.F., et al. (2004) Pushing the envelope: challenges in a frame-based representation of human anatomy. Data Knowl. Eng., 48, 335–359[CrossRef].
Parker, S.P. Synopsis and Classification of Living Organisms, (1982) McGraw Hill.
Rogers, J. and Rector, A. (2000) GALEN's model of parts and wholes: experience and comparisons. Annual Fall Symposium of American Medical Informatics Association, , LA , pp. 714–718.
Rosse, C., et al. (1998) Motivation and organizational principles for anatomical knowledge representation: the digital anatomist symbolic knowledge base. J. Am. Med. Inform. Assoc., 5, 17–40
Rosse, C. and Mejino, J.L. (2003) A reference ontology for biomedical informatics: the foundational model of anatomy. Biomed. Inform., 36, 478–500.
Schulz, S. and Hahn, U. (2004) Towards a paradigm for biomedical structure. Proceedings of KR-Med, 63–71.
Smith, B., Williams, J., Schulze-Kremer, S. (2003) The ontology of the gene ontology. Proceedings AMIA Annual Symposium 2003 , pp. 609–613.
Smith, B. and Rosse, C. (2004) The role of foundational relations in the alignment of biomedical ontologies. Proceedings of Medinfo 2004, , pp. 444–448.
Winston, M.E., et al. (1987) A taxonomy of part–whole relations. Cognitive Sci., 11, 417–444.
Wos, L. The Automation of Reasoning: An Experimenter's Notebook with Otter Tutorial, (1996) Academic Press.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||