Ontological modeling at a domain interface: bridging clinical and biomolecular knowledge

The work presented in this paper reflects the initial activities in the NEUROWEB project (project number 518513). We wish to thank the clinical partners: Istituto Nazionale Neurologico Carlo Besta (INNCB, Milan, Italy), Orszagos Pszichiatriai esNeurologiai Intezet (AOK-OPNI, Budapest, Hungary), University of Patras (UOP, Patras, Greece), Erasmus Universitair Medisch Centrum Rotterdam (MI-EMC, Rotterdam, Holland). In particular, we wish to thank Dr Yiannis Ellul and Dr Stella Marousi from UOP, Dr Zoltan Nagy and Dr Csaba Ovary from AOK-OPNI, Dr Aad Van Der Lugt and Dr Philip Homburg from MI-EMC, and Dr Giorgio Boncoraglio from INNCB, for the valuable contributions during the knowledge acquisition campaign and model refinement process.

This example was discussed during the tutorial of the Bio-ontologies Special Interest Groups (SIG), within the joint bioinformatics conference ISMB/ECCB (Intelligent Systems for Molecular Biology/European Conference on Computational Biology) 2007, in Wien.

Following (Laudan, 1977), we refer to the notion of Research Tradition as a specialization of Kuhn’s paradigm concept. While a paradigm casts a general light over science and its methods, relying on a historical and cultural background (e.g. Western Medicine versus Eastern Medicine), a Research Tradition better represents a specific corpus of meta-theories (e.g. Atomism, Materialism) and methodologies (e.g. Inductivism, Operativism) acting in the same paradigm.

By methodology we mean the corpus of methods (i.e. rules and principles) guiding the sensory and cognitive processes required to achieve the goals of a specific practice. Whereas in AI it is possible to clearly distinguish between a content theory and a mechanism theory (Chandrasekaran et al., 1999), human knowledge displays circular relations between the two components: for instance, the use of different ‘sensory devices’, and different mechanisms to elaborate their results, is normative for any ontological categorization. Although ontological modeling addresses the content theory component, we argue that the methodological aspects should be taken into account, especially when dealing with tacit knowledge.

International guidelines should not be regarded as exhaustive formulations of diagnostic procedures, but rather as minimal requirements to be met in order to formulate correct diagnoses.

A Research Community can be epistemologically conceived as the crossing over between a scientific community that adheres to a hypothetical-deductive paradigm of scientific thinking, and a Community of Practice (Brown & Duguid, 1991; Wenger, 1998; Hildreth et al., 2000) that does not strictly adhere to it. Whereas hypothetical-deductive knowledge is often encoded in formal constructs (e.g. in physics, Newton’s Law of Mechanics; in Chemistry, the Periodic Table), the knowledge of a Community of Practice tends to be implicit, that is, tacit knowledge (Nonaka & Takeuchi, 1995), or encoded as a semi-formal Representational Artifact.

We acknowledge that certain schools of ontological engineering advocate the representation of ‘reality’, avoiding to mingle with epistemological complications (Smith, 2004). The theoretical debate between Conceptualism and Realism is a long-standing issue in philosophy; realism advocates the principle of identity as ‘objective in its essence’; conceptualism argues for the existence of conceptualizations of reality, shared across different subjects and communities. It is out of the scopes for this paper to address that debate, and rule out the ideal approach for ontological modeling. Nonetheless, we argue that the refusal to recognize the differences between conceptualizations, and their underlying rationale, is a major weakness of the realistic approach. Assuming a generally valid reality, as in the case of general-purpose ontologies, may also neglect the specific elements of the expert knowledge accrued within a Research Community.

In the Semantic Web community, this problem would be typically identified as a Data Integration problem (Benassi et al., 2004). Here we specifically stress the methodological implications of divergence, hence the need of a knowledge-accurate ontological model to support the integration task.

Specifically, the diagnostic process should not be regarded as a sum of independent tests on single parameters, but rather as the progressive recognition of the patient’s state, through mutual reinforcement of different observations; that means that a decision tree is not a suitable model for the cognitive processes underlying the diagnostic activity.

As we will display with more details in the next session, these logical formulas are formally encoded as DL axioms.

As these relations express the partonomy of the phenotype (i.e. its constitutive parts), they can be regarded as specialized mereological relations (cf. Simons, 1987; Sattler, 2000).

We adopt the term ‘Biological’ Process instead of ‘Pathophysiological’ in analogy to the Gene Ontology Biological Processes.

The Atherosclerotic Stroke implies an obstructive event in the large brain arteries; the Cardioembolic Stroke implies the generation of the obstructing body (thromboembolism) in the heart; the Lacunar Stroke implies an obstructive event in the small brain arteries.

The only case in which this correspondence does not occur is when the CDS indicator has a coarse anatomical resolution, though more details on the anatomical part can be referred to the Low Phenotype, as in the case of the Small Vessel Disease.

Other approaches, such as the one adopted in OQAFMA (Mork et al., 2003), define SQL-like languages (StruQL in OQAFMA) to support queries over semantic descriptions. The disadvantage of these approaches is the need to learn a specific query language, which can reduce the usability of the system by medical users, who are not already familiar with writing SQL queries.