HCLS face to face day 2 -- 21 Oct 2008

Outreach with Karen Myers

<ericP> scribenick: Susie

Discusses potential outreach opportunities

Potential outreach HCLS event in December in NJ

CSHALS

Luncheon at Bio-IT

Semantic Technologies Conference was popular this year

WWW2009 and ISWC2009 are also events to watch out for

InfoTechPharma is another opportunity

<marie> hi

Discuss conferences

<marie> for the record - the INRIA workshop on 28 Nov. 08:

<marie> http://www.inria.fr/actualites/colloques/2008/lscc2008/index.en.html

-> http://esw.w3.org/topic/HCLSIG/Conferences Prioritized conferences

Top Five Conferences

<ericP> 1 HIMS Feb Orlando

<ericP> Krisine

<ericP> AMIA Summit Nov 9 DC

<ericP> Krisine

<ericP> AMIA trans-med bio-informatics feb, SF

<ericP> Susie (conditional on acceptance of an already submitted proposal)

<ericP> 3 ISMB June Stockholm

<ericP> Marco, Paul, Andrew, Jun, Matthias

<ericP> Bosse

<ericP> 4 BioIT World April BOS

<ericP> Susie

<ericP> 5 DIA March Berlin

<ericP> Susie

<ericP> marie, i note that euroBio didn't make it on the top five, but do you think you of ivan will attend?

Karen asks people to blog

Karen asks for volunteers who can speak with the media,

<marie> not sure eric. I'll check the potential content for the 2009 edition, and then we'll decide if it's worthwhile

Karen reviews lists of journals

Marie-Claire proposed we publish special editions that highlight our work

Karen discusses reaching out to multiple audience, e.g. hc vs pharma; CIO vs specialist

Karen encourages us to publish, as it'll help with storytelling

Need a high level story

Need story for people in health care

Health care can be different in

different parts of the world

<SusieS> Discussion of possible health care storytelling

<ericP> http://esw.w3.org/topic/HCLSIG/Media

<Kristin> Using Semantic Web Technologies for Neuroimaging - Christine Golbreich

<Kristin> scribenick

<Kristin> THe problem is that the content of images is not accessible to computers

<Kristin> Adding annotations (metadata) to the images -

<Kristin> We also need to add meanings to these annotations

<Kristin> These need controlled vocabulary used to define the meanings and agree on the meanings

<Kristin> A specific example is labelling of the sulci (fissures) and gyri (gray matter)

<Kristin> They need this to be done automatically as much as possible

<Kristin> Ontology of Sulco-Gyral anatomy

<Kristin> They have agreed on a controlled vocab. Based on work by Dameron.

<Kristin> More details on this: http://www.irisa.fr/visages/pdf/AMIA2008-AMCGXMBG.pdf

<Kristin> They used OWL DL

<Kristin> There are classes that follow all the Gyri and Succi anatomy -

<Kristin> The ontology they developed has some specificity - shared organizational memory - reference ontology for their specific application with specialized vocabulary

<Kristin> For instance, patches, lesions, separations, and segments.

<Kristin> They vocabulary represent the standard anatomy ontology--theirs decorating the lower leaves in the hierarchy to describe the parts they need.

<Kristin> They would like to have takening an existing one and extract what was needed. But there were some difficulties in doing this.

<Kristin> DL expressivity - SHIF (d)

<Kristin> They have 300 classes

<Kristin> They have 30 object properties

<Kristin> There are three main steps:

<Kristin> First segmentation of the images--definition of patches

<Karen> Susie: Discussion about what other standards organizations HCLS should liaise with

<Karen> This afternoon presentation from HL7

<Karen> ...main standards organization for healthcare

<Karen> ...CDSC for clinical trials

<Karen> ...what about bioinformatics?

<Karen> Helen: Open EHR

<ericP> openEHR home page

<Karen> Kristin: some countries don't have same abilit

<Karen> ...CDC is working on a global Open EHR

<Karen> ...systems that don't work unless you have Soc. Security number

<Karen> ...CDC looks at this because of SARS...global health issues

<Karen> Susie: Let's look at what organizations are doing, other new ones coming out

<Karen> ...see what they are all doing

<Karen> ...engage them earlier

<Karen> ...more opportunities to be influential

<Karen> ...possibly DICOM

<Karen> ...digital images

<Karen> ...could be an interesting group

<Karen> ...wonder about scientific realm

<Karen> ...such as groups working on XML expessions of genes

<Karen> ...one area the minimum information people

<Karen> ...about annotations

<Karen> ...not an expert here, but would love your input

<Karen> ...about the critical organizations

<Karen> Scott: suprised to learn that Affymetrix has released

<Karen> ...Go annotations of ?

<Karen> Marco: through projects and organizations you find these people

<Karen> Susie: EBI; human protein project

<Karen> Marco: they have their own ML

<Karen> Scott: Microray standards evolve toward proteomics platforms

<Karen> ...imaging aspects

<Karen> ...sets of experiments

<Karen> Marco: It's heterogeneous in bioinformatics

<Karen> ...what do you want to get out of this?

<Karen> Susie: Define things in a more consistent way

<Karen> ...and do queries

<Karen> ...so if things defined in databases were somewhat similar

<Karen> Marco: Maybe there is something for that

<Karen> ...recent meeting I mentioned not just exporting

<Karen> ...synchronizing to XML format for proteomics data

<Karen> ...may consider going to a different level

<Karen> ...not really ready for that message

<Karen> ...some work to be done there

<Karen> Eric: So they are defining XML standards? [yes]

<Karen> ...so if you take someone who has to come up with a message format

<Karen> ...and say use this, show some Turtle

<Karen> ...if you show not in level of this is what SW can do

<Karen> ...but rather as this is a way to express things

<Karen> ...then get them to express it textually

<Karen> ...and start to realize what they can do with SW later on

<Karen> ...show bits of the benefit, not swollow whole pill

<Karen> ...explain why we like it

<Karen> Marco: I'm easy to convince

<Karen> ...a few people may want to go to that

<Karen> ...go to trend in that community to go with ML

<Karen> ..may be odd at first in that community

<Karen> Eric: Landscape they are in, people put their IP into DTB, or XML Schema

<Karen> ...rather than RDF

<Karen> Susie: Let's hold discussion for later on strategic planning

<Karen> Scott: What's going on

<Karen> ...general sense in community for need to have not only agreements

<Karen> ...about users of URIs and people who want to use SW

<Karen> ...what are best ways to mint URI if you don't have one

<Karen> ...an effort to align on that level

<Karen> ...discussion on mailing list of HCLS

<Karen> ...also from data provider POV

<Karen> ...imagine if you named a protein; Uniprot provides you wit standard name

<Karen> ...following linking open data principles,

<Karen> ...agree with data providers such as Uniprot

<Karen> ...notion is to get a few data providers on board

<Karen> ...to serve up meaningful RDF

<Karen> ...and boot strap people to use common URIs

<Karen> ...so that we can more easily link up

<Karen> ...not link up step by step

<Karen> ...Workshop coming up

<Karen> ...National Center for Bio-ontology

<Karen> ...hope to get data providers and data users to this workshop

<Karen> ...suggestion to provide official PERLs for HCLS community

<Karen> ...So we have had some discussions

<Karen> ...social networking has been done

<Karen> ...by Eric for a while now

<Karen> ...spoke in Viennea

<Karen> s/Eric Neumann

<Karen> s/Vienna

<Karen> ...EN has been trying to drum this up

<Karen> ...our colleagues at Science Commons

<Karen> ...also working on this issue

<Karen> ...get the user side lined up.

<Karen> ...Communities like MyGrid, BioMoby

<Karen> ...agree on a common naming scheme

<Karen> ...before we agree on details

<Karen> ...get started with a PERL server at NCBO

<Karen> ...show data providers with examples of how to deliver RDF versions of data

<Karen> Susie: Good idea to get key players together

<Karen> ...hope to do early next year; hope it will happen

<Karen> ...hope for NIH backing

<Karen> Scott: Invite you to get beind this

<Karen> ...still some negotiation going on

<Karen> ...who wears what hats

<Karen> ...being addressed

<Karen> Q: Why not?

<Karen> [?]

<Karen> Scott: OKKAM

<ericP> okkam home page

<Karen> ...European project to produce URIs for all things semantic

<Karen> ...to manage

<Karen> ...they can help

<Karen> ...as HCLS we provided a use to them

<Karen> Susie: Sound like good people to be at workshop

<Karen> ...be back at 2:00pm for Charlie Mead, HL7 presentation

<Kristin> Charlie Mead, NCI - HL7

<Kristin> Name based on OSI stack

<Kristin> Had a pretty rapid uptake because it was the "only game in town" for transferring information

<Kristin> Increasing adoptin lead to questions about the other problems it could solve (e.g., accounting, orders mgmt..)

<Kristin> Orgainzational structure of HL7 organized by domain experts (by not necessarily medical professionals)

<Kristin> Interop happened through virally, not but formalisms.

<Kristin> HL7 has become the victim of its own success.

<Kristin> It worked well in hospitals because of syntactic compatibility.

<Kristin> "HL7 isn't a standard, but a style guide."

<Kristin> The system was brittle.

<Kristin> Other countries started showing up at HL7 meetings because they were starting to think about it.

<Kristin> --> 1995-2006 Success Drives New Approaches ...3.0 was to solve the problems of 2.x

<Kristin> It used a shared reference information model (pre-UML)

<Kristin> Because HL7 was built by engineering types, they felt they should stay out of terminology.

<Kristin> They realized that they really need to think about semantic interop

<Kristin> v3.0 has a 1% adoption because it is too hard.

<Kristin> Application rules--what type of apps is it? what type of messaging?

<Kristin> It is still being built bottom up even though the efforts were strong to drive it top down

<Kristin> The four pillars of sematnic interop came out in the process of developing HL7

<Kristin> They are necessary, but not sufficient

<Kristin> #1: common-model (or harmonized sibling models) across all domains of interest

<Kristin> #2: Have to have an abstract data type specification

<Kristin> ... there is now a shared ISO data specification

<Kristin> #3: Must be a way for bindng terms from concept based technologies

<Kristin> #4: (most important result in v3) A formally defined process for defining specific structures to be exchanged across machines (a data exchange standard)

<Kristin> Five backbone concepts (plus role-relationships)

<Kristin> Entity, rule, participation, act and act relationships

<Kristin> This is a real rich deep model

<Kristin> roles not rules

<Kristin> These ideas existed, they just came up with a standard representation.

<Kristin> Everyone knew about it, but they were all using different models

<Kristin> All of these classes are state machines--you have to manage state, but everyone does it differently.

<Kristin> Collection, context and attribution - building complex RIM-based structures

<Kristin> You can form arbitrarily complex structures - semantically equivalent - but not computational equivalent because of the nature of XML

<Kristin> Information and terminology model...

<Kristin> ..intersecting and interleaving semantic structures

<Kristin> There are common structures for domain semantics and common structures for terminology so we should be able to bind htem together.

<Kristin> But this crashed.

<Kristin> These different terminologies have different levels of complexity

<Kristin> These turned out to be semantically interoperable in the XML model

<Kristin> People are now starting to take a look at how this can be fixed through OWL and RDF

<Kristin> HL7's Clinical Document Architecture -

<Kristin> ...emerged coicidentally with the development of XML

<Kristin> Driven by the document-centricity of much of the healthcare practise -

<Kristin> six fundemental document characteristics: persistence, authentication, stewartship, wholeness, global/local content, human readability

<Kristin> This didn't take off in the US,but it did take off in Europe because they didn't have the legacy 2.X versions in place

<Kristin> Version 2 is seeing more adoption in the US

<Kristin> CDA surfaced something that HL7 was never able to do - semantic interop

<Kristin> CDA allows for highly informational systems to interact with less information systems in an organized way

<Kristin> How does HL7 pplay into HCLS

<Kristin> It wa focused on doctors and hospitals

<Kristin> up to 1997

<Kristin> CDC showed up, veternary, CDISC, FDA all started to show up...

<Kristin> This meant they needed to branch out to cover more than clinical care

<Kristin> Of particular interest to W3C HCLS is RCRIM TC

<Kristin> This is the immediate match between HCLSIG and HL7

<Kristin> NCI is funding "Transcend" out o fUCSF

<Kristin> Tolven - at UCSF

<Kristin> The lack of a behavior model in v3 is impacting people who have adopted HL7 (for instance National Health Service in UK)

<Kristin> HL7 when through a reorg

<Kristin> Went from an unstructured org to a very structured organization with a CEO (MD from Intel), board of directors, CTO, technical steering committee

<Kristin> ..create an architecture board.

<Kristin> Starting to look like an org that can solve this problem

<Kristin> HL7 is collaborative and actively pursuing collaborations inside and outside the US

<Kristin> Best example of collaboration is CDISC - gone on the longest and had the most success (not all success,but a good deal of them)

<Kristin> Programmers talking to care providers caused more of a problem because they weren't using the same languages

<Kristin> gave more details and got further apart from each other

<Kristin> A level of abstraction was useful

<Kristin> BRIDG - (CDISC - HL7) a way to have abstractions (analysis model) to enable interactions and not get lost in the XML details

<Kristin> The good news about BRIDG has been rock solid for four years

<Kristin> The problem is that there are subdomains.

<Kristin> People want to see 'their words'

<Kristin> e.g., observations instead of adverse effects--one set of users wants to see their labels, not an abstraction

<Kristin> So now there is a multi-layered model

<Kristin> OWL and HL7

<Kristin> Mapping domains of interest descri bed by OWL and visualized by visual conceptionalization

<Kristin> ...Currently these are not directly connected

<Kristin> The took the BRIDG model in UML

<Kristin> For each concept attribute pair

<Kristin> They ran this through prodigy

<Kristin> This discovered the errors, conflicts, etc.

<Kristin> This provided more robustness

<Kristin> Service awareness in HL7

<Kristin> ..initial work began in 2006 - health services specification proejct

<Kristin> most of this was produced outside HL7

<Kristin> AFter thinking about it, they determined what was missing

<Kristin> They came up with the emerging architecture framework

<Kristin> SAFE (sp?)

<Kristin> SAEAF

<Kristin> intersecting concepts from SOA, RM for ODP

<Kristin> The Reference Model of Open Distributed Processing, intersecting with HL7, and service oriented architecture

<Kristin> Using CDL at the NCI - pi4soa

<Kristin> Business process modeling version 2

<Kristin> Choreography based view that can be mapped into the HL7 space

<Kristin> Fixes the dynamic modeling problem

<Kristin> Now HL7 and the W3C can have a more meaningful discussion

<Kristin> Clinical research filtered query - CRFQ

<Kristin> Two situations:

<Kristin> 1. ERH repository

<Kristin> ...clinical trial cohorts are recruited through advertising

<Kristin> ...submit through a service interface to get qualified patients

<Kristin> 2. You are a patient

<Kristin> ...or a doctor

<Kristin> ...you have to know if there is a clinical trial you are qualified for

<Kristin> knowledge sharing is hoping to play a part in this

<Kristin> caBIg is hoping to help with the new pharma model

<Kristin> They believe knowledge sharing will play a part

<Kristin> NCI is working in a "BIG-Health" initiative

<Kristin> HL7's role in these two contexts -

<Kristin> 1. Bringing key components

<Kristin> 2. More adoption

<Kristin> 3. better collaboration

<marco> q: What is the formal representation of HL7?

<marco> a: it is in UML and the java code, but not something formal like OWL-DL

<marco> Back to the ARFQ - Charlie really sees a need for this.

<marco> ... there is a business model connecting patients with clinical trials

<marco> ..helps pharma and the patients

<marco> Pharma is willing to pay for the data in the clinical repositories

<marco> ...this is better than current model for patient recruitment

<marco> People in other countries don't have a patient ID

<marco> q: what about privacy--how do you deal with it?

<marco> a: HL7 doesn't define standards--there is a group that comes together to work on it

<marco> NCI has a lot of security built into it

<marco> This has not been well addressed.

<marco> Partly because of a lack of agreement on patient ids

<marco> EricP Q: The notion of security they (NCI) preserve is role based

OWL2 Update [Christine Golbreich]

<ericP> scribenick: ericP

zakplease dial riviera_b

Christine: [slide: what's new in OWL2?]
... OWL2 defines several profiles
... these are used to define subsets that have certain performance/expressivity charactersistics
... [slide: increased expressivity power]
... e.g. MaxCardinality(3 boundTpo Hydrogen)
... other addition, of utmost importance to us: property chain inclusion axioms
... e.g. uncle rule
... property chains and qualified cardinality were needed for expressing SNOMED
... added {Reflexive, Irreflexive, Asymmetric}
... note that Irreflexive states that X cannot be a *proper* part of itself
... added local reflexivity
... added disjoint properties
... (increases expressivity)
... added DisjointUnion
... syntactic sugar for having unions applied to disjoints
... added DisjointClasses (syntactic sugar)
... extended datatypes
... + numbers taken from XSD
... + strings with a language tag
... + boolean values, binary data, uris, tim instances, ...
... added data type restrictions
... added simple metamodelling based on punning
... cannot use the same name for a class and a datatype
... added databse style keys
... defined fragments beyond OWL Lite
... e.g. OWL 2 EL:
... .. useful for large number of properties or classes, e.g. SNOMED CT or NCI thesourus
... .OWL 2 QL, for large a box sets
... .. can do query answering while the data to remain in the DB
... OWL 2 RL, which is implementable in DBs with rules

<DanC_lap> ericp, an IFP applies to _all_ classes. Keys can (a) have more than one column, and (b) be restricted to a class

Q's about transitivity and differences between keys and IFPs

PLING Update [Renaldo]

Renaldo: meeting this week between PLING and PrimeLife to addressprivacy in social networks
... examining implications of policy langes like CC (and more expressive ones) when applied to social data
... many policy langs leaves interop probs
... [PAW goals]
... looking for use cases from HCLS

Marco: will this help social network sites integrate?

Renaldo: [enthusiasitic "yes"]
... e.g. mvoing data and policies from linkIn to MySpace

Marco: for example, MyExperiment would like blogging avail to its users without providing the site itself

helenC: in health care, hppa is a typical use case

Renaldo: is that solved with machine-processable code?

[No]

henryS: do you need to move into data licensing?

Renaldo: there's an open data group trying to use CC-like licesnese for data

[Alexander]

SWIG Update [Alexander]

danbri: started as rdfig
... as an IG, we wouldn't have any work, except you folks spoil the IG expectations
... we don't go to other folks and hide the ugly SW bits
... we can learn from HCLS there

Alexandre: [XSPARQL]
... can we access RDF and XML with one language?
... SPARQL CONSTRUCT produces SPARQL query results which are processable in XML
... can query RDF with XQuery if you use a normalized RDF expression in XML
... will submit XQPARQL to W3C?

henryS: is there a library I can use now?

Alexandre: when we do the submission, we expect to publish the library

mscottm: is perfomrance a motivation?

Alexandre: not really, more more to get XML files

henryS: is this the perfect GRDDL transform language?

RIF update [Christian de Sainte Marie ]

Christian: everyone understands data interchange
... RIF is designed to bridge between different rule languages
... zillions of existing rule languages
... superset approach: create a language that encompasses 80% of the existing rule languages
... doesn't work
... intersection approach: express intersection of these languages
... we couldn't do wthis without knowing the needs for more expressivity
... so we started with two dialects and looked for their intersection
... we have a catalog of expressions
... maybe when RIF is deployed and weell-used, perhaps folks will migrte to it
... you don't need to look at the dialects for a rule set
... striped model is easily machine processable
... BLD has single literal conseequents (HORN-like)
... BLD symbol space divided into predicates, functions, ...
... built-ins are functions or predicates which are specififed elsewhere
... won't tell you about named function arguments 'cause i want them out
... [slide: UML-ish element hierarchy]
... in theory, you can check a RIF document against that picture
... [slide: PRD Overview]
... PRD conditions are like BLD conditions - logic functions, + NAF
... it looks like there is a core, but it's not clear yet that there is a usable core
... charter ends May 09, so hope to finish by then

s/May/July/ # ?

scribe: not sure if UC&R is on Rec Track

henry: do you have named graph constructs?
... contextual logic, like "if X believes Y than I do too"

Christian: if you use "believes" as a standard (first order) relation then yes
... if you are talking about 2nd order, we have not constructs for that
... we don't even have negation because this is where rule languages split

henryS: for example i could query graphs wtih SPARQL

Christian: if you use an external function, in theory, yes

ivan: have there been discussions about PRD with OWL/RDF content?
... e.g. I could model as NOT as a removal from the RDF triple store

Christian: this is an objective
... to mesh BLD with RDF
... the priority on the PRD side is to import data in the form of XML schemas

ericP: what's the theoretical vs. mechanical balance?

Christian: we just have an interchange goal. if everyone has a different notion of negation, then there is no interchange

mscottm: 2 questions and a memory stick:
... .. what about DLP and being able to program with DL?
... .. and what about fuzzy rules?

Christian: we explicitly excluded designing yet another modeling language

mscottm: i meant interop between OWL and RIF

Christian: yes, it's in the SWC document

<mscottm> OWL- RL (can be implemented with a set of rules)

ivan: there's a stronger connection than that: the OWL 2 RL can be implemented with rules which can be written in BLD
... Dave Reynolds hasd been checking whether OWL 2 RL can be expressed in BLD

Christian: Dave Reynolds hasd been checking whether OWL 2 RL can be expressed in BLD
... outcome seems to be yes
... re: fuzzy rules, shouldn't be a problem, depending on where you want the fuzziness

harryH: why would RIF bee good for the KIF camp?
... and why would SWIG channel switch from using n3 rules?

Christian: basic argument is "do i want to interchange it without thinking about what the remote side uses?"
... maybe down the line we'll see editors for RIF
... am interested in how much of Common Logic BLD covers

helenC: your BLD is like the RDF for our data

Christian: the idea is that you have an implementation of BLD is not a rule engine; it's a translator
... we did tests on top of JS and production rule languages

wrap up

Susie: i think brains are full, so let's have a drink outside

ADJOURNED

- DRAFT -

HCLS face to face day 2

21 Oct 2008

Attendees

Contents