BAG TECHNIQUE Bag technique-a tool making use of public health bag through wh...
Formal Ontology Meets Industry: Best Practices
1. On some best practices in large-scale
ontology development:
The Chronious Ontology Suite
as a Case Study
Luc Schneider[1], Mathias Brochhausen[1],
David Koepsell [2]
[1] Institute for Formal Ontology and Medical Information
Science, Saarland University, Saarbrücken, Germany
[2] Technical University Delft, Delft, Netherlands
FOMI, TU Delft, Delft, July 7-8, 2011
2. The importance of best practices in
ontology development
• Methodological pluralism in ontology developments reflects
differing challenges and facilitates solutions adapted to the
respective business requirements.
• It may lead to ad hoc selections of categorical criteria or
design choices and the adoption of kludges, as opposed to a
consistent adherence to foundational principles. Ad hoc
solutions may cause impediments in the development and the
usage of products.
• Applied ontologies based on sound design principles and
theoretical foundations provide advantages compared to
ontologies constructed in an ad hoc manner, namely
consistency, re-usability and harmonisation.
3. The CHRONIOUS Ontology Suite
• The CHRONIOUS Ontology Suite has been developed in the context
of a European project aiming at the development of an integrated
telemedical platform for monitoring the general health status of
patients with chronic health conditions and providing decision
support for clinicians.
• The CHRONIOUS Ontology Suite consists of three components:
– the Middle Layer Ontology for Clinical Care (MLOCC)
– the COPD (Chronic Obstructive Pulmonary Disease) ontology
– the CKD (Chronic Kidney Disease) ontology.
• MLOCC contains 476 classes, the COPD ontology 964 and the CKD
ontology 972.
4. Outline of the talk
• We illustrate a number of best practices in large-
scale ontology construction by using examples drawn
from the CHRONIOUS Ontology Suite.
• We will focus on three topics:
– realist upper level ontologies as a means for ex-ante
harmonisation
– modularization of ontologies and reuse of pre-existing
resources
– general design principles to ensure the consistency of
taxonomies.
6. The rationale for realist top-level
ontologies (1)
• To guarantee data exchange across heterogeneous sources,
we must resort to ontologies that unify the different ways in
which the domain is represented by the various end-users and
information systems designers.
• Realist ontologies, i.e. ontologies that depict reality
independently of the mental or digital representation of
reality by end users and knowledge engineers, provide a
unified way of representing the domain from the start,
without the need of an ex-post integration of heterogeneous
perspectives.
7. The rationale for realist top-level
ontologies (2)
• Upper level ontologies help to adjust a system/service to an
ever growing user-base and to access a growing number of
heterogeneous data repositories.
• Upper level ontology development profits from the adoption
of a realist view, since the representation of reality by the
ontologies allows all possible perspectives.
8. Basic Formal Ontology
• Basic Formal Ontology (BFO) grows out of a philosophical
orientation which overlaps with that of DOLCE and SUMO.
Unlike these, however, it is narrowly focused on the task of
providing a genuine upper ontology which can be used in
support of domain ontologies developed for scientific
research, as for example in biomedicine within the framework
of the OBO Foundry.
• Thus BFO does not contain physical, chemical, biological or
other terms which would properly fall within the special
sciences domains. BFO is the upper level ontology upon which
OBO Foundry ontologies are built.
10. MLOCC as an extension of BFO
• The Middle Layer Ontology for Clinical Care (MLOCC) and thus the COPD
and CKD ontologies have been built on top of BFO by appending the
upper-level classes of MLOCC onto leaves of BFO. By linking MLOCC-
classes to BFO-classes, the meaning of the latter are given a reality-driven
semantics.
– The BFO-class Disposition subsumes the MLOCC-classes Disease and
Malfunction.
– The BFO-class Object subsumes MLOCC-classes such as Organism,
Chemical Substance, Institution and TechnicalObject (including devices
and instruments).
– The BFO-class Process subsumes the MLOCC-classes Intentional
Process and Natural Process; the first subsumes classes related to
human and social activities, in particular medical (diagnostic and
therapeutic) processes, while the second subsumes Chemical Process
and Organismal Process.
12. Avoiding design mistakes using
upper-level ontologies
• A realist upper-level ontology provides constraints on the
classification of domain entities. These restrictions prevent
ontological mistakes that can lead to costly re-designs of
domain ontologies. Some examples from BFO/MLOCC are:
– Occurrents do not participate in other occurrents. A process of
heart beating does not undergo an increase, but a heart rate,
which is the quality of an organism resulting from a heart
beating, does.
– Dispositions, functions or roles do not participate in occurents,
but are realised by processes. The renal filtration function does
not change, but is realised by the process of renal filtration
which has a glomerular filtration rate as an outcome.
14. Perspectivalism, modularity and
reusability
• Perspectivalism is the recognition that there are many
representations of reality that are equally adequate because
they capture different salient aspects of the same world.
• Reality can be assayed
– in terms of substances and their qualities or powers as well as in
terms of processes;
– at various levels of granularity, ranging from the atomic and
molecular levels to those of cells, tissues and organisms.
• Hence, reality cannot be accounted for in terms of a single
monolithic ontology, but only in terms of a multitude of
modular ontologies that are orthogonal to each other and
thus are also re-usable.
15. Modularity of the CHRONIOUS
Ontology Suite (1)
• The CHRONIOUS ontologies are built on top of an established
upper-level ontology, namely BFO, as far as classes are
concerned.
• As to relations or object properties, we have used the
Relations Ontology (RO) of the OBO Foundry, a set of formal
relations that are used in biomedical applications. Aside from
minor modifcations, the object properties of the CHRONIOUS
ontologies represent an extension of RO.
• The branch below the class Organismal Independent
Continuant mirrors the structure and content of the
Foundational Model of Anatomy (FMA), a reference ontology
for the domain of anatomy
18. Modularity of the CHRONIOUS
Ontology Suite (2)
• The core of the Middle Layer Ontology for Clinical Care has
been extracted from the ACGT Master Ontology.
• Moreover, the COPD Ontology and the CKD Ontology import
MLOCC, which represents the common core of the chronic
disease ontologies, that expand on it in domain-specific ways.
Modular Structure of the
CHRONIOUS Ontology Suite
20. Formal constraints on taxonomies (1)
• Taxonomies should contain only types, not instances or
tokens.
– However, information objects like questionnaires are
ontologically tricky in this respect. Is Chronic Respiratory Disease
Questionnaire a type or a token ?
• Taxonomies are based on formal subsumption, i.e.
subsumption ties have to be rigid and context-independent.
– „Tiotropium bromide is a bronchodilator drug“ has to be
interpreted as „a certain amount of tiotropium bromide is used
as a bronchodilator“. To be a bronchodilator is a a role, i.e. an
extrinsic feature of tiotropium bromide. Hence Tiotropium
Bromide is not a subclass of Bronchodilator (and hence Drug),
but a Chemical Substance that has the role of a Bronchodilator.
21. Formal constraints on taxonomies (2)
• Multiple inheritance of primitive classes should be avoided.
– TiotropiumBromide cannt be subsumed both by Bronchodilator
and ChemicalSubstance. Instead of multiple inheritance, one
should privilege intrinsic or formal subsumption on the one
hand, and role attribution on the other.
• Sibling classes should be disjoint.
• UnknownX as well as other catch-all classes for remaining
cases should be avoided. Such classes do not cut at a joint of
reality, i.e. do not represent an ontological, but an
epistemological distinction.
22. Conclusion
• We have illustrated some best practices related to
– ex-ante harmonisation through realist upper-level ontologies,
– modularisation and re-use
– general guidelines pertaining to the consistency of taxonomies.
• Our source of examples is the CHRONIOUS ontology suite, a
large-scale bio-medical ontology development project.
• From our experience we can vouch for the efficiency of the
strategies and principles described above for stream-lining
ontology construction and ensuring the quality and adequacy
of domain ontologies.
23. Acknowledgements
Research leading up to the present article has been supported by
the ICT-2007-1-216461 grant within the Seventh Framework
Programme of the EU, as well as by a post-doc grant from the
National Research Fund, Luxembourg (cofunded under the Marie
Curie Actions of the European Commission [FP7-COFUND]), and
has been carried out under subcontract to the Fraunhofer
Institute for Biomedical Engineering, St. Ingbert (Germany).