RDF and non-RDF Databases on the Semantic Web

W3C Semantic Web Projects.

Eric Prud'hommeaux, Sanitation Engineer.
Last modified: $Date: 2009/01/21 14:33:31 $

This work is licensed under a Creative Commons Attribution 3.0 License, with attribution to W3C.

Use Case - Clinical Studies

Pharmaceuticals pay bundles to test drugs.
Pharmaceuticals express protocol in CDISC.
--precipitous gap--
Hospitals exchange info in HL7/RIM ontologies
Hospitals have relational databases

SemWeb Windfalls

Web-scale data integration.
Standardized reasoning.

Standardized Reasoning

RDFS

spot a dog.
dog subClassOf mammal.
mammal subClassOf animal.

OWL
```
X an animal.
X hasChild Y.
Y a kitten.
```

Drug Mapping

X rdfs:subClassOf set1.
set1 a owl:restriction.
set1 owl:onProperty may_treat.
set1 owl:someValuesFrom   :C0028754
=> X a weightLossDrug.

Merged Data Ideal

Many DBs contributing
to a network of data.

Databases as RDF

Observation O
on Pation P
by Doctor D.

Query Interface

Observation O
on Pation P
by Doctor D.

Triples Transformation

Warning: vapid example:

Observation O
on Pation P
who has Doctor D.

Triples Transformation

(Virtual) Transformation Pipe

Query the World

Configured query services
federating and distributing
to many DB Interfaces.

Translation layer

SWObjects - SPARQL CONSTRUCT.
d2r - RDF schema and conf in RDF.
Virtuoso - DDL.
squirrelRDF, et al, - ???.

CONSTRUCT Expressivity

   PREFIX :mydb <http://cityhospital.example/dbs>
CONSTRUCT { ?o a          :PatientObservation .
            ?o :patient   ?p .
            ?o :doctor    ?d }
            ?d :foaf:name ?dName }

    WHERE { ?o mydb:patient ?p .
            ?o mydb:doctor  ?d .
            ?d mydb:name    ?dName }

Diabetic Patient Data

Patients
Tests
Prescriptions

SDTM Query

PREFIX sdtm: <http://www.sdtm.org/vocabulary#>
PREFIX spl: <http://www.hl7.org/v3ballot/xml/infrastructure/vocabulary/vocabulary#>

SELECT ?patient ?dob ?sex ?takes ?indicDate # ?indicEnd ?contra
 WHERE {
  ?patient a sdtm:Patient ;
          sdtm:middleName ?middleName ;
          sdtm:dateTimeOfBirth ?dob ;
          sdtm:sex ?sex .

  ?ip	  sdtm:subject ?patient ;
	  sdtm:standardizedMedicationName ?takes ;

	  # nothing by ingredient in sdtm/caBIG/...
#          sdtm:startDateTimeOfMedication ?indicDate ;
	  # ; sdtm:endDateTimeOfMedication ?indicEnd
	  spl:activeIngredient ?ic . ?ic spl:classCode 6809
   .
  OPTIONAL {
  ?cp	  sdtm:subject ?patient ;
	  sdtm:standardizedMedicationName ?contra ;
# 	  sdtm:startDateTimeOfMedication ?contraDate ;
	  # ; sdtm:endDateTimeOfMedication   ?contraEnd
	  spl:activeIngredient ?cc . ?cc spl:classCode 11289
   .
  }
  FILTER (!BOUND(?cp))
} LIMIT 30

Pipeline

  cat sdtm.rq | \
  SWtransformer -q - hl7-sdtm.rq | \
  SWtransformer -q - db-hl7.rq -s http://hospital.example/DB/ | \
  mysql -u root DiabeticPatientsDataSet --table

HL7 Query

Conspicuously Absent

Node renaming - <http://cityhospital.example/dbs/patient?id=12817>.
Rule interaction - has limited expressivity.
transitivity must be forward-chained in the DB

Performance

Performance winfalls are hard to achieve:

push queries to remote DB (pushing queries down).
occasional winfalls by cashing (conjested sources).

Kudos

thanks to Lilly and Lincoln Labs for funding this work.