Re: Implications and Issues in Converting UML to RDF/OWL When Constructing HL7 RIM Ontology

Hi Alan,
Here is the use case if you are missing it:


Dan Russler, M.D.
VP Clinical Informatics
Oracle
(VM) 404-439-5983
--- Original Message ---

Helen, Dan, Vipul, Bill


The only way to be certain the technical solution you propose "has legs" is to work through SEVERAL QUALITATIVELY DISTINCT examples of the Use Cases driving the creation of the system.  This is even more true for semantically oriented computational infrastructures than it is for those built around a fixed information model. 


First.  I seem to have come in in the middle and can't find the beginning.  What's your use case for representing HL7 in OWL?  


We've been working on representing the RIM  - and RMIMs, DMIMs, etc, in OWL for some time now with Tom Marley (cc'ed above). The focus has been on capturing the constraints both in the UML diagrams and in the text boxes that surround them, as well as on the binding of various coding systems, particularly SNOMED, to HL7.  


Tom is just finishing up an ITS in OWL that covers the basic RIM including all the data types and hopes to have that for the HL7 meeting in a week's time.  I am revising a paper for publication on the Terminology Binding Interface, and will send it along as soon as I've dealt with the referees' comments.  (It will be presented at KR-MED and published in Applied Ontology).   I'll also be at one of the pre-meetings around Hl7 but not at HL7 itself because of other commitments.  I am working on a more comprehensive paper on the methodology 


In the course of this we have done a lot of thinking and experimenting on the relationships between UML and related OO modelling paradigms and OWL.  You are probably familiar with the OMG meta model, which is helpful, but highly generic.  I would be cautious about a generic solution to a specific problem such as the RIM.  Also the RIM has a number of idosyncracies, that need attention, some related to current controversies between Tom Beale and Gunther Schadow. 




We look forward to future discussions.  Please keep us in the loop.


Regards


Alan


...


A few commenbts on Helen's excellent summary: 




1. Mapping UML classes to RDF classes 
 UML class hierarchy is shown in class diagram.  In a typical OO fashion, if class B is the subClass of class A, all attributes of A will be "inherited" by B.  Furthermore, class B often will have additional attributes that impose further restrictions on B.  RIM class diagram only shows those subclasses that have additional attributes compare to their super-classes.  Translation of RIM class diagram into RDF/OWL classes and subclasses can be quite misleading.   I find a more adequate categorization is the class type list in the domain vocabulary, for example, for classification of the act class, I have used  the domain ActClass ("A code specifying the major type of Act that this act-instance represents" ). 


The RIM classes only provide a starting point for an OWL model
In the OWL mapping there end up being many more classes.  If there are constraints between, for exmple,  ActClass and other things, then one way to express them is to construct a subclass defined by that ActClass value, and then add the constraints.   




2. Mapping UML attributes RDF properties 
In OO, all attributes of a class will be inherited by its subclasses.  In RDF/OWL, there is not such "inheritance", such that 


{?A ?P ?X. ?B rdfs:subClassOf ?A} => {?B ?P ?X}   


(the above rule does not exist in OWL semantics) 


Some semantics in OO inheritance can be mapped to RDF/OWL using restrictions, but great care must be applied not to over-restrict your ontology.  Maybe we can look at some details during one of our Tcon. 


If I understand you, this is part of a recurring theme in OWL models.  Attachment of slots/attributes has no precise OWL equivalent.  Perhaps the best way to express it is as minCardinality 0 which has no effect on semantics but can on tools.  A necessary but not sufficient condition is that the topic ?B be in the domain of the property ?P, so a common convention is to add ?"B as a disjunct to the domain of ?P.   However, this really imputes both more and less semantics than is appropriate. 


One also has to be very careful of the OWL open world assumption and lack of a unique name assumption.  If the purpose of the owl model is to validate instances, then it has to be necessary to close the OWL model and the instances.  There are several ways to do this, but the easiest is to reify attributes as classes.  Then it is easy to close the list of attributes when that is needed for validation.  The same goes for associations, although since there are fewer the problem is different. 


There are also problems with some of the elaborate HL7 data types. 


3. Mapping "relationship classes" to RDF property 


There are "relationship classes" in RIM, such as ActRelationship class, RoleLink class.    These classes are used to state relationship between two acts, or two roles.  In OO, they are designed as classes, and can be easily mapped to the ER model for database, in order to accommodate  one-to-many relationship.  In RDF/OWL, it is natural to model them as properties, and their types as sub-properties.  This is how they are modelled in our RIM ontology mentioned in my previous email. 


I would be cautious about representing ActRelationship etc. as a property because they have additional attributes.  If those attributes matter, then the only way to represent them is to represent the relationship as a class.  The RIM does a great deal of reification of classes.  We have followed it pretty literally in our work, but I think it is a significant design issue to reconsider which things should be reified systematically.  ("Reify" in this usage means to represent a relationship as a class rather than a property so that it can be the topic of additional restrictions.  See the n-ary relations note at Semantic Web Best Practice and Deployment WP (SWBP). http://www.w3.org/TR/swbp-n-aryRelations/








These are a few points I have encountered during my exercise of constructing RIM Ontology.   


I am copying HCLS list on this subject, hope to hear from other people's experience in "converting RIM ontology from UML" or from UML->OWL in other domains.  Many domain knowledge is captured in UML type of models and diagrams. If we can find a good way to convert these domain knowledge in UML to OWL, I believe it would offer some value for Informatics as general. 










I


On 27 Aug 2006, at 23:55, DAN.RUSSLER@ORACLE.COM wrote:


Hi Michael...This looks like a well-considered discussion. Thanks for the reference!

Dan Russler, M.D.
VP Clinical Informatics
Oracle
(VM) 404-439-5983
--- Original Message ---
>Hi All,

There is there Ontology Definition Metamodal specification that is an soon to be adopted specification from the OMG that was put together by IBM and Sandpiper (http://www.omg.org/docs/ad/06-05-01.pdf) that provides quite a bit of insight into the relationship between UML and OWL and provides mappings at different levels between the two.

cheers,
Michael

Michael Miller 
Lead Software Developer 
Rosetta Biosoftware Business Unit 
www.rosettabio.com 
-----Original Message-----
From: public-semweb-lifesci-request@w3.org [mailto:public-semweb-lifesci-request@w3.org] On Behalf Of William Bug
Sent: Sunday, August 27, 2006 11:18 AM
To: DAN.RUSSLER@ORACLE.COM
Cc: Kashyap, Vipul; helen.chen@agfa.com; public-semweb-lifesci@w3.org
Subject: Re: Implications and Issues in Converting UML to RDF/OWL When Constructing HL7 RIM Ontology


Hi Dan, 


This a very helpful foundation to start from, Dan.  I would suggest selecting 4 - 5 very specific examples of this use case, preferably covering across the examples of family medical history clinical statements including a variety of different semantic entities - e.g., genetic information, documented treatments of relatives for disorders that include a genetic component, environmental conditions including life style info and socio-economic status, etc..  Would it be possible to build on the stroke victim Use Case Helen et al. have assembled on the Wiki (http://esw.w3.org/topic/HclsigDscussionTopics/HclsSubGroupACPP)?


I would also give an strong endorsement of of Vipul's statement "we need to be very careful about translating from the UML to the OWL meta-model.  It is quite likely that there are multiple alternatives in OWL for representing the same UML construct and we would probably need best practices for the same."


One thing is for certain, "definitely require" should be substituted for "probably need" in the quote above.


I would also add the following: 


A) UML is an immensely useful formalism for creating or communicating to others a given computable information model.  Unfortunately, despite the many similarities between the process by which a formal OOP class-based model designed to compute on real-world artifacts and the process by which one constructs an formal ontological framework to characterize real-world entities, these two orientations toward creating practical, computational infrastructures have many conflicting requirements.  
B) As an example, the specific nature of UML-style modeling created to support OOP system development and - in some circumstances - interoperability (between implementation languages and persistence mechanisms, for instance) tend to have properties that are orthogonal to the goals of ontology development and ontology application.  This is even true when one considers the mechanisms provided in UML and/or XML for extending and combining models.
C) The general issue when it comes to developing semantically mediated systems interoperability (such as would be required for the Use Case you describe here, Dan) is the dual goal of designing a mechanism and consequently implementing an infrastructure both widely applicable to the problem domain AND effective to the required level of semantic granularity on any specific example case can also often conflict with one another.  The only way to be certain the technical solution you propose "has legs" is to work through SEVERAL QUALITATIVELY DISTINCT examples of the Use Cases driving the creation of the system.  This is even more true for semantically oriented computational infrastructures than it is for those built around a fixed information model. 


For a helpful overview of how these issues hold for HL7 RIM, I'd recommend reviewing the following:
1) HL7 RIM: An Incoherent Standard, Smith, B and Ceusters, W. (August 2006)
http://ontology.buffalo.edu/HL7/doublestandards.pdf


2) Methods in biomedical ontology., Yu, A.C. (2006), J Biomed Inform. 2006 Jun;39(3):252-66
http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=pubmed&cmd=Retrieve&dopt=AbstractPlus&list_uids=16387553&query_hl=2&itool=pubmed_docsum
http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6WHD-4HRN3GB-1&_coverDate=06%2F30%2F2006&_alid=439843079&_rdoc=1&_fmt=&_orig=search&_qd=1&_cdi=6848&_sort=d&view=c&_acct=C000007158&_version=1&_urlVersion=0&_userid=95578&md5=8befcae03811ddf4a1101c027f020dc5
http://www.sciencedirect.com/science?_ob=MImg&_imagekey=B6WHD-4HRN3GB-1-W&_cdi=6848&_user=95578&_orig=search&_coverDate=06%2F30%2F2006&_qd=1&_sk=999609996&view=c&_alid=439843079&_rdoc=1&wchp=dGLbVzz-zSkzS&md5=bcd20c1e022879d25c7d5f70e4cc4707&ie=/sdarticle.pdf


The former is a well-informed review of the conflicting HL7 RIM objectives of providing BOTH an information model AND a reference ontology.  It's a bit stern in tone, but it's purpose is very much constructive in nature.  I'm not an expert in the use of HL7-RIM, so I'd most definitely defer to others on this list to provide counter arguments to the issues they raise most specifically as relates to 'B' and 'C' above.  The basic recommendation is to simply separate the data/info modeling from the reference ontology features into two distinct efforts, analogous to the way FuGE and FuGO approach this issue for functional genomics information.


The later is an excellent overview providing a clear sense of how and why to avoid conflating data modeling with ontological engineering.  It only very briefly touches on HL7-RIM, but the entirety of the review is extremely relevant to this topic.


I don't mean to appear overly didactic here.  Quite the opposite - working with folks on this list has been an immensely valuable learning experience for me.  I just thought it would be helpful to provide a brief overview of the specifics on this issue Vipul has raised in the context of HL7-RIM.


Cheers,
Bill




On Aug 27, 2006, at 11:11 AM, DAN.RUSSLER@ORACLE.COM wrote:


Hi Vipul,
Excellent comments...It seems to me that an ontology that doesn't have a testable use case can never be judged useful or not useful for any kind of predictable search strategy.
Here is the use case I would like to explore:
"An RDF system has been put in place to navigate to healthcare resources stored in many systems including genetic resources, proteomic resources, and healthcare medical record resources. One of the healthcare resources is a CDA document, which itself has a URI. However, within the CDA document, are detailed resources expressed as HL7 RIM Acts which also carry URI's. I would like to navigate from my RDF-based navigation system to a specific kind of HL7 RIM Act, e.g. a family medical history clinical statement, within a specific kind of CDA document. "
I'm sure this use case could be better written with a little time. However, is this the kind of use case you were looking for?

Dan Russler, M.D.
VP Clinical Informatics
Oracle
(VM) 404-439-5983
--- Original Message ---

> I agree with Helen. In general, we need to be very careful about translating from the UML to the OWL meta-model.
It is quite likely that there are multiple alternatives in OWL for representing the same UML construct and we would
probably need best practices for the same.
Towards this end, it would be very useful to understand the use cases and requirements for which this translation is required.
The questions we need to answer are:
-          What is the value of this translation? May be for the particular use case and requirement, it?s not really useful?
-          What are the assumptions behind the use cases? Typically these assumptions would help determine a more appropriate 
and accurate translation.
RIM appears to be more of a ?meta-model? rather than an ?ontology? and the semantics of the various constructs have not been
defined accurately. There has been a lot of discussion. Also, one needs to understand issues related to whether we are building an information model
for HL7 message content or an information model for persistence to be mapped into a database schema or one for decision support?
I would strongly recommend that this exercise, which is very valuable, should be carried out in the context of a well defined use
case. Doing it in the abstract could result in making arbitrary modeling decisions which may detract from the usefulness of the
translation.
Cheers,
---Vipul



From: public-semweb-lifesci-request@w3.org [mailto:public-semweb-lifesci-request@w3.org] On Behalf Of helen.chen@agfa.com
Sent: Friday, September 08, 2006 5:13 PM
To: DAN.RUSSLER@ORACLE.COM
Cc: public-semweb-lifesci@w3.org
Subject: Implications and Issues in Converting UML to RDF/OWL When Constructing HL7 RIM Ontology 

Hi, Dan 

Given the detailed description of HL7 RIM and Domain Vocabulary, it is quite tempting to generate the RIM ontology directly from some kind of conversion scheme between UML and RDF/OWL , preferably some automated tools/scripts.   

I assume you must aware some similar efforts in generating RIM ontology. One such ontology is made by Bhavna Orgun (http://www.ics.mq.edu.au/~borgun/Software.html).   

However, when examining closely the fundamental theories behind UML (OO) and RDF/OWL (Model theory), you will find that the direct translation from UML to RDF/OWL is not so straightforward as it seems.  For example: 

1. Mapping UML classes to RDF classes 
 UML class hierarchy is shown in class diagram.  In a typical OO fashion, if class B is the subClass of class A, all attributes of A will be "inherited" by B.  Furthermore, class B often will have additional attributes that impose further restrictions on B.  RIM class diagram only shows those subclasses that have additional attributes compare to their super-classes.  Translation of RIM class diagram into RDF/OWL classes and subclasses can be quite misleading.   I find a more adequate categorization is the class type list in the domain vocabulary, for example, for classification of the act class, I have used  the domain ActClass ("A code specifying the major type of Act that this act-instance represents" ). 

2. Mapping UML attributes RDF properties 
In OO, all attributes of a class will be inherited by its subclasses.  In RDF/OWL, there is not such "inheritance", such that 

{?A ?P ?X. ?B rdfs:subClassOf ?A} => {?B ?P ?X}   

(the above rule does not exist in OWL semantics) 

Some semantics in OO inheritance can be mapped to RDF/OWL using restrictions, but great care must be applied not to over-restrict your ontology.  Maybe we can look at some details during one of our Tcon. 

3. Mapping "relationship classes" to RDF property 

There are "relationship classes" in RIM, such as ActRelationship class, RoleLink class.    These classes are used to state relationship between two acts, or two roles.  In OO, they are designed as classes, and can be easily mapped to the ER model for database, in order to accommodate  one-to-many relationship.  In RDF/OWL, it is natural to model them as properties, and their types as sub-properties.  This is how they are modelled in our RIM ontology mentioned in my previous email. 

These are a few points I have encountered during my exercise of constructing RIM Ontology.   

I am copying HCLS list on this subject, hope to hear from other people's experience in "converting RIM ontology from UML" or from UML->OWL in other domains.  Many domain knowledge is captured in UML type of models and diagrams. If we can find a good way to convert these domain knowledge in UML to OWL, I believe it would offer some value for Informatics as general. 

Kind regards. 

Helen 







http://www.agfa.com/w3c/hchen



DAN.RUSSLER@ORACLE.COM 
08/25/2006 12:34 PM ToHelen Chen/AMPJB/AGFA@AGFA 
cc
SubjectRE: invitation for next ACPP call








Thanks for the welcome Helen. I hope to learn a lot and perhaps, at least, contribute a little. 
I've been exploring transforms from UML to RDF and have found some resources. Do you have anything written on mappings or transforms from the RIM to RDF expressions?

Dan Russler, M.D.
VP Clinical Informatics
Oracle
(VM) 404-439-5983
--- Original Message --- 

Hi, Dan and the group 

My sincere apology to all with regard to this week's ACPP Tcon.   I was on vacation the whole week and did plan to attend the Tcon on Tuesday.  Due to a health problem of my father, I ended up only getting back to my computer this morning. 

Welcome, Dan, to this group.  It is so good to hear your interest in working on HL7 RIM ontology.  We have the same intention and worked on a draft which you can find at ACPP group's attachments section: 

http://esw.w3.org/topic/HCLS/ACPPTaskForce?action=AttachFile 

You can download the RIMV3OWL.zip file which contains all files of RIM ontology in protege.    I had some design consideration and choices made during the design and implementation of this ontology and discussed them once with Chimizie.  I am very much looking forward to hearing your comments and working together. 

I hope to talk with all of you on our next Tcon at 3pm, Tuesday, August 29.   

Kind regards. 

Helen 
http://www.agfa.com/w3c/hchen



"Davide Zaccagnini" <davide@landcglobal.com> 
08/22/2006 08:27 AM To<DAN.RUSSLER@ORACLE.COM>, <ogbujic@bio.ri.ccf.org>, <thongsermeier@partners.org>, Helen Chen/AMPJB/AGFA@AGFA, <Alfredo.Morales@cerebra.com>, "'Brandt, Sam \(MED US\)'" <sam.brandt@siemens.com> 
cc
SubjectRE: invitation for next ACPP call









Dan,

Here are the details 

SW_HCLS(ACPP) 
SW Life Sciences IG 
Tuesdays 
3:00pm-4:15pm
+1.617.761.6200, conference code 2277 ("ACPP")

Regards

Davide

-----Original Message-----
From: DAN.RUSSLER@ORACLE.COM [mailto:DAN.RUSSLER@ORACLE.COM] 
Sent: Monday, August 21, 2006 6:18 PM
To: ogbujic@bio.ri.ccf.org; thongsermeier@partners.org; helen.chen@agfa.com;
Alfredo.Morales@cerebra.com; davide@landcglobal.com; Brandt, Sam (MED US)
Subject: Re: invitation for next ACPP call

Thanks!...I'm looking forward to working with this group.

Can someone send call-in information?


Dan Russler, M.D.
VP Clinical Informatics
Oracle
(VM) 404-439-5983

--- Original Message ---
> 
> I second that!
> Welcome Dan!!
> 
> -Sam
> 
> 
> -----Original Message-----
> From: Hongsermeier, Tonya M.,M.D. <THONGSERMEIER@PARTNERS.ORG>
> To: Davide Zaccagnini <davide@landcglobal.com>; helen.chen@agfa.com
> <helen.chen@agfa.com>; DAN.RUSSLER@ORACLE.COM <DAN.RUSSLER@ORACLE.COM>;
> Chimezie Ogbuji <ogbujic@bio.ri.ccf.org>; Alfredo Morales
> <Alfredo.Morales@cerebra.com>; Brandt, Sam (MED US)
> Sent: Mon Aug 21 17:54:34 2006
> Subject: RE: invitation for next ACPP call
> 
> Davide
> 
> This is fantastic, Dan, welcome to the team
> 
> Best
> 
> Tonya
> 
>  
> 
> Tonya Hongsermeier, MD, MBA 
> Corporate Manager, 
> Clinical Knowledge Management and Decision Support 
> 
> Partners HealthCare System 
> Clinical Informatics Research and Development 
> 93 Worcester Street, PO Box 81905 
> Wellesley, MA 02481 
> P: 781.416.9219     Mobile: 617.717.8711 
> Fax: 781.416.8912  
> 
> ________________________________
> 
> From: Davide Zaccagnini [mailto:davide@landcglobal.com] 
> Sent: Friday, August 18, 2006 3:00 PM
> To: helen.chen@agfa.com; DAN.RUSSLER@ORACLE.COM; Hongsermeier,
> Tonya M.,M.D.; Chimezie Ogbuji; Alfredo Morales; sam.brandt@siemens.com
> Subject: invitation for next ACPP call
> 
>  
> 
> Hi All,
> 
>  
> 
> I would like to invite to our next ACPP call Dan Russler, VP
> of clinical informatics at Oracle. Dan is interested in
> approaches to map HL7 RIM to RDF and in modeling clinical
> protocols with semantic technologies. 
> 
>  
> 
> Dan, 
> 
> here is the link to our wiki page where we can start looking
> at the work we have done so far
http://esw.w3.org/topic/HCLS/ACPPTaskForce#preview
> 
>  
> 
> Are we ok for next Tuesday at 3pm?
> 
>  
> 
> Regards
> 
>  
> 
> Davide 
> 
>  
> 
>  
> 
>  
> 
> Davide Zaccagnini, MD, MS
> 
> Product Manager, Medical Informatician
> 
> + 1 617 864 1031 (Office)
> 
> + 1 617 642 7472 (Cell)
> 
> davide@landcglobal.com
> 
> www.landcglobal.com
> 
>  
> 
> 
>
----------------------------------------------------------------------------
---
> This message and any included attachments are from Siemens
> Medical Solutions 
> USA, Inc. and are intended only for the addressee(s).  
> The information contained herein may include trade secrets or
> privileged or 
> otherwise confidential information.  Unauthorized review,
> forwarding, printing, 
> copying, distributing, or using such information is strictly
> prohibited and may 
> be unlawful.  If you received this message in error, or have
> reason to believe 
> you are not authorized to receive it, please promptly delete
> this message and 
> notify the sender by e-mail with a copy to
Central.SecurityOffice@siemens.com
> 
> 
> Thank you


> 




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 - William.Bug@DrexelMed.edu








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.




-----------------------
Alan Rector
Professor of Medical Informatics
School of Computer Science
University of Manchester
Manchester M13 9PL, UK
TEL +44 (0) 161 275 6149/6188
FAX +44 (0) 161 275 6204
www.cs.man.ac.uk/mig
www.clinical-esciences.org
www.co-ode.org



>

Received on Monday, 28 August 2006 19:13:27 UTC