Re: [HCLS] Bridging Ontologies - with Foundational Ontologies

[William Bug]

Hi All, I think Chimezie and Matthias are definitely on the right path here. As I see it, the bottom-up approach to semantic KE/KR that SemWebTech is so suited to must be wedded to the top-down approach using an upper level ontology such as DOLCE or OBR (version of BFO designed for use in biology**). As Matthias and Chemezie have both pointed out, if you have two data repositories covering overlapping domains of knowledge, if they don't use the same fundamental formalized _expression_ for those core entities they have in common - say, for instance, a continuant (or perdurant) - an entity whose existence extends over some time interval - then it won't be possible to automatically determine which entities they describe are in fact related in a fundamental way.  I most definitely agree with them based on the work we are doing on the BIRN Ontology Task Force.  We are trying to import the FuGO ontology - recently expanded to cover all biomedical experimental domains at a "generic" level and to be re-christened - likely to Ontology of Biomedical Investigation (OBI) [don't quote me on that].  We need to make certain the collection of "Biomaterial Entities" we specifically define for annotating data in BIRN are ontologically equivalent at the base level.  The only way for us to do this is for OBI and the BIRNLex knowledge representation framework we are building to both import and reference the same upper level ontology.  Unfortunately, at this time, there is no published version of UBO (see below), so we both end up recapitulating the upper level ontology elements in our individual OWL files.  Ultimately - and hopefully very soon - there will be a UBO for us both to reference externally. If you don't approach the task of semantic interoperability with a shared ontological graph where all the nodes are very clearly defined according to a set of empirically derived "best practices" (e.g., the OBO Foundry Principles - http://obofoundry.org/) - say you were to do a single "bridge" between NeuronDB and CocoDat - there is no guarantee that bridge will be of any use beyond linking NeuronDB to CocoDat.  In other words, it won't necessarily follow the mapping can link either NeuronDB or CocoDat to any other knowledge repository covering overlapping domains - even if you use terms from the UMLS.  As we discussed before - and as Olivier from the NLM UMLS research staff mentioned - UMLS is not an ontology, it is a unified thesaurus of the medical lexicon as described by several dozen popular, overlapping controlled vocabularies.  There are all sorts of ontology construction "best practices" it's underlying semantic graphs do not obey (e.g., do not combine distinct mereotopological "part of relations" and subsuptive "is a" relations all in the same graph).  This is even true of the UMLS Semantic Network - the closest thing UMLS has to an ontology.  Olivier mentioned work is underway to review the SN and consider how to bring it closer in line with ontology construction "best practices" without breaking existing systems (http://mor.nlm.nih.gov/snw/). I would add I don't see this approach as being incompatible with bottom-up approaches to KE/KR.  Along the lines Vipul mentioned in response to Don, it certainly is possible to refine and improve the underlying semantic graph based on observed statistical correlations amongst related entities.  One can also put complex network dependencies in place with weights adjusted according to periodic reduction of the accruing semantically formalized repository. At this point in time, it difficult to imagine how the underlying ontology could auto-adjust to such feedback.  An ontology is a bit less flexible that the sort of classification and summarization methods used by Google - no matter how sophisticated the algorithmic implementation of such a system may be.  However, periodic review could certainly help point out emerging logical inconsistencies (and novel consistencies).  Still, devising a means to "evolve" the ontological graph based on statistical analysis of semantically formal representation of phenotype observations (and commensurate formal descriptions of the investigative procedures used to make those observations) is really what will provide the greatest payoff, I believe.  It will help tie together several important threads: - deriving a deeper understanding of the emergent higher level activities such as neurodevelopment, learning, memory, and cognitive processing TO genetic networks, enzymatic pathways, receptor-mediated cellular events, and ion-channel triggered cell physiological events; - linking observations made in animal models to clinical practice. Cheers, Bill **OBR (http://ontology.buffalo.edu/medo/biomedo.htm) will likely be re-christened UBO and be associated with the OBO Foundry effort.  In fact, use of UBO will likely become one of the OBO Foundry "Principles" On Aug 22, 2006, at 5:11 AM, Matthias Samwald wrote: I also think that smaller foundational ontologies like DOLCE are the key to interoperability between ontologies. It seems like the only way to ensure good interoperability is the agreement on some basic structures and design patterns. When two OWL ontologies are based on two totally different world-views to start with, it becomes almost impossible to build a bridge between the two that serves some practical use. Maybe we should look more into the use of foundational ontologies and best-practices of their use and extension, and less on the alignment of ontologies that have already developed into totally different directions In metaphoric words: Our task is not to build brigdes between stray villages on different islands. Our task is to find the most livable island and tell all of the villagers to settle there.  Unfortunately, bio-zen was more oriented to concepts at the  molecular  level than for my needs, but any connection made to DOLCE would  automatically orient themselves with bio-zen (or any other  derivatives). Great to hear that! It really seems that most of the promises of semantic web ontologies are only realised when top-level ontologies like DOLCE are used. Maybe we should evaluate the potential use of DOLCE or BFO for the BioRDF tasks? kind regards, Matthias Samwald Bill Bug Senior Research Analyst/Ontological Engineer Laboratory for Bioimaging  & Anatomical Informatics www.neuroterrain.org Department of Neurobiology & Anatomy Drexel University College of Medicine 2900 Queen Lane Philadelphia, PA    19129 215 991 8430 (ph) 610 457 0443 (mobile) 215 843 9367 (fax) Please Note: I now have a new email - [EMAIL PROTECTED] This email and any accompanying attachments are confidential. This information is intended solely for the use of the individual to whom it is addressed. Any review, disclosure, copying, distribution, or use of this email communication by others is strictly prohibited. If you are not the intended recipient please notify us immediately by returning this message to the sender and delete all copies. Thank you for your cooperation.