Ontology alignment for wearable devices and bioinformatics in professional health care

Tracking #: 826-2036

Jack Hodges
Mareike Kritzler
Florian Michahelles
Stefan Lueder
Erik Wilde

Responsible editor: 
Guest Editors Ontology and Linked Data Matching

Submission type: 
Application Report
Web Ontology Language (OWL) based models and triple stores hold great potential for access to structured information. Not only are OWL-based ontologies extremely versatile and extendable, but triple stores are robust against changes to ontologies and data. The biomedical field illustrates this value insomuch as it employs vast amounts of information distributed across different models and repositories. This paper presents a case study that sought to demonstrate the real-world value of linking disease, symptom, and anatomical models with wearable devices and physical property models and repositories. Integrating these models is both necessary and problematic; necessary to provide undifferentiated access to health care professionals, problematic because although the biomedical ontologies and repositories exist, they aren't semantically aligned and their designs make alignment difficult. This case study demonstrated that manually linking multiple biomedically-related models can produce a useful tool. It also demonstrated specific issues with aligning curated ontologies, specifically the need for compatible ontology design methodologies to ease the alignment. Although this study used manual ontology mapping, it is believed that systems can be developed that can work in tandem with subject matter experts to reduce mapping effort to verification and validity checking.
Full PDF Version: 


Solicited Reviews:
Click to Expand/Collapse
Review #1
Anonymous submitted on 08/Nov/2014
Review Comment:

This paper describes a case study for manual ontology alignment in the context of wearable devices and bioinformatics in professional health care. The authors mainly discusses a possible strategy which ontologies could be linked to support physicians in locating wearable devices for a patient having a specific diagnosis. Their aim is to show the value of semantic integration by showing how medical professionals could benefit by having integrated access to biomedical models/repositories.

I agree in that overcoming semantic heterogeneities e.g. by aligning ontologies has not been addressed in all domains and further work is necessary. For instance, in the medical field there are still many open issues.

However this application paper does neither show how anyone can profit of this work nor gives a convincing evaluation of this matching problem. What is the concrete result of this work? How could any physician profit of the few detected correspondences? A manual alignment was only realized for four diseases. There is no discussion who concretely needs this case study - was there some “real world” motivation, e.g. due to a problem in a real collaboration with some hospital? If so – it is not mentioned in this paper.

Currently, a concrete alignment strategy of the five selected ontologies DOID, FMA, SYMO, QUDT and SSN is not described. The authors discuss semantic bridges that can assign a “high level” semantic type for the alignment that will be determined between the ontology concepts. Some given examples do not highlight the issues of aligning the concepts. For instance, there is a description of a manual web search strategy but I feel this is not “a solution to anything really problematic”. It is probably the standard technique for a medical expert ..

It is unclear what the authors mean with “data” – do you mean instances? What are examples of data, how did you use the data? Are their real annotations between data and ontological concepts?

I propose to combine this work with some existing automatic approach and do an evaluation using the manual alignment. Semantic bridges could be used by an automatic matching approach. It needs to be shown that concrete correspondences between all five ontologies have been detected, e.g. by showing the number of identified correspondences and measuring the quality of the results (precision, recall, fmeasure, logical coherence ...).

Overal the currently descibed methodology is poor looking at the huge amount of related work in this area. Moreover, the publication neither gives an evaluation nor shows a real application in some project nor gives user access to the methodology and/or results, such that it is not adequate for publication in SWJ.

Some specific comments:

In the abstract, triple stores and the importance of owl are mentioned. Why? Both are not used.. The sentence “Not only are OWL-based ontologies extremely versatile and extendable, but triple stores are robust against changes to ontologies and data.“ must be removed since there is no connection to the paper content at all.

The authors define some goals, that seem to be not achieved (?) but at least not discussed (!). Moreover, the four goals are not concrete (why four? what is the content of each goal?). I would prefer to highlight few real contributions in the introduction.

Related work:
Data and information integration have been studied intensively - a.o. there has been a lot of research on ontology matching/alignment in the last 15 years. In particular, a lot of approaches have been published describing domain-related methodologies, e.g. for the biomedical field. Some surveys on ontology mapping and alignment give an overview to common approaches, e.g. the book “Ontology matching” (Euzenat, Shvaiko, 2007). In the last years, the life science tracks (Anatomy & LargeBio) of OAEI (Ontology Alignment Evaluation Initiative) showed how successful several systems are in automatically aligning anatomy and other biomedical ontologies (AML, Logmap, GOMMA …) [http://oaei.ontologymatching.org/]. The authors discuss one interesting approach of Rance et al. suitable for their domain. However there are further approaches that could be used and discussed in this context, e.g. “Instance-based matching of large life science ontologies” (Kirsten et al 2007) or “Alignment of biomedical ontologies using life science literature” (Tan, 2006). Such approaches could be used in this paper, e.g. the authors mentioned some manual use of a search engine -> instead the search could be done automatically and results could be combined with some standard linguistic matching.
Overall, the current discussion of related work is poor, i.e. existing literature in field of ontology alignment needs to be discussed appropriately!!

What does this mean?:
“In each case, since the original models are curated by separate entities, new mapping models (aka semantic bridge ontologies [8,5] were created and then merged with the original..)” --> this is confusing. There is research on ontology merging, which is absolutely not trivial! What are mapping models? The herein used terms are confusing to some extent looking at existing work in the field of schema and ontology alignment /matching / mapping / merging etc.

Minor comments:
The corresponding ontologies and concept identifiers (accessions) should be shown in all illustrations, e.g. the reader might not know that “body part” is an FMA concept.
“but it wasn’t being used” - -> why do you mention?

currated --> curated

Review #2
By Clement Jonquet submitted on 28/Nov/2014
Review Comment:

The papers presents an application of semantic web approach for integrating health/medical data with wearable device information. The idea is to offer an integrated model for: device, physical property, Anatomy, Symptom and disease. The authors explain the ontologies they have used and how they have connected them.
The paper has significant issues to be considered good candidate for a journal publication. Mainly, the study presented in this ‘application report’ appears not finished, with negative conclusions in terms of scalability/expansion of the approach and impact.
Although the position of the study is never clear between data integration vs. data interoperability, the paper does present an integrated model for 5 ontologies: DOID, SYMP, FMA, QUDT and SSN. Such models is supposed to help to represent device related data (at least Vandrico) and answer queries such as the one given in beginning of introduction. Such a model is an interesting contribution which certainly has value and should be evaluated by the community.
However, the biggest lack is that the model proposed is not really experimented/tested/used in the paper. Its impact is not demonstrated. By the end of the paper, 4 diseases (from DOID) are mentioned, but no information about the numbers of concepts/relations from other ontologies are given. The methodology for generating manually symptoms for disease is purely manual and such a task should not be given to non-medical expert. And if medical experts would be in the loop, they would know the disease-symptoms relations without having to search manually the web. In addition, as stated by the authors this approach totally prevent to be extended for more diseases and syndromes.
Overall in the paper the clarification about what the authors called ‘alignment’, ‘mapping’, ‘semantic bridge’ etc. is not clear. The notion of alignment is pretty clear in the community and I am not sure I will call ‘connecting ontologies to integrate them in a common schema’ an ontology alignment. What you are doing is a good example of designing a new small schema or ontology with strong reuse of other existing ones. Which is a good practice, but it’s not ‘ontology alignment’ rather ontology reuse.
Such point would have be avoided if the paper would provide a real state of the art related to: the use of wearable device and relevant ontologies for them and previous work that have proposed an integration with biomedical ontologies. This is a strong lack of the current paper.
In conclusion, I will say that the current paper does not offer convincing evidence of the impact and importance of the application. The core of the contribution (i.e., the integrated model) might be useful (assuming it does not exists, what a state of the art on that aspect would have said) but the application of that model does not convince the reader of the results one can obtain by using such a model. Semantic web technologies are used at least by the fact of offering the new model as an ontology. But nothing related to semantic web data technologies is mentioned (eg., RDF etc.).

Major comments by sections:
- Abstract: “undifferentiated … professionnals”. = unclear
- Abstract: “a useful tool” : such what?
- Use of section numbering and structure is obscure. Unique subsections are used.
- Section 1: You should discuss that the query the doctor is asking in the case of diabetes II will be asked only once… then the doctor will have the knowledge that diabete => deviceX.
- Section 1 should rather concentrates on wearable device rather than on the impact of ontologies and semantic web technologies. The audience of the SWJ will know this.
- Beginning of section 2 is unclear. Your goals are described with words that haven’t been clarify to the reader yet. Maybe come back on this in conclusion.
- If you assume a device is always something that measure a property, then say it explicitly.
- You could give examples in beginning of section 2 to illustrate your speech.
- Section 3.1: explain what you mean by semantic bridge. If your contribution is a “semantic bridge ontology” define this introduction, give it a name and refer to it by its name.
- Use namespace abbreviation in your figure, this will help figuring out what is existing, what is yours. Provide your own namespace.
- You need to tell us more about Vandrico data source. Size, format, importance in the field, why this one, etc.
- Section 4 is not a relevant state of the art for your application. This section must allow to answer what have been done in the semantic web for medical and wearable device integration? Nothing on mapping (also it is not necessary if you don’t call your work mapping/alignment anymore). Nothing on device.
- Mission conclusion that comes back on the contributions and discuss them before detailing the perspectives.
Minor comments:
- Section 1: “locate” : do you mean “find out”
- Section 2: ‘4’ => four
- ‘Figure 1:’ => ‘Figure 1.’
- Section 2.1 exists without section 2.2. Idem for 3.1
- Section 2.1: ‘OBO Disease Ontology’ => don’t need OBO. Idem after.
- Fig 2 is important in your paper but totally unreadable.
- Section 3.1 ‘be exist’ => English
- Fig 3 is also too small.