An RDF Data Model for the Semantic Web

5th Oracle Life Sciences User Group meetingMay 16-17, 2005

Agenda

Introduction – 5 min– Susie Stephens

Semantic Web for Life Sciences – 25 min– Susie Stephens

Oracle support of RDF in RDBMS – 25 min– Souripriya Das

Demo of Siderean’s Seamark Navigation Server – 25 min– Mike DiLascio, David LaVigna & Joanne Luciano

Discussion – 10 min– Susie Stephens

Semantic Web for Life Sciences

Susie Stephens

What is the Semantic Web?A machine-readable format that is Web compatibleThe Semantic Web adds definition tags to information in Web pages

– Enables computers to discover data more effectively

– Allows new associations to form between pieces of information

Resource Description FrameworkW3C standard for the common data formatBased on triples (subject–predicate–object)Everything has a URI Ontologies used to label the RDF tagged elements

Image Source: W3C

Image Source: W3C

Enterprise Integration Hub

Image Source: W3C

Semantic Web Stack

Image Source: W3C

Pharma Productivity

Source: PhRMA & FDA 2003

Critical Path Initiative

Source: Innovation or Stagnation, FDA Report, March 2004

Ontology Frameworks for Integration

<translatesTo>

mRNA <located>

Localization

<targets>

<affectedTissue>

Intervention point<MOA>

<probeFor>

Drug

Protein

<participatesIn>

<partOf>

Cascade pathway

Bio-process

<drugInteraction>

Target model

<affecting>

Treatment

Gene

<influences>

Disease

<profiledBy>

Microarray experiment

<efficacyMarkerFor>

<hasProduct>

<transcribes>

Biological Pathways

Image Source: Cytoscape

Beyond the “Dead” Graphical Model

Image Source: KEGG

Assigning Trust Values to Data

Image Source: SWANS

InferencingIf Gene G is implicated in Disease D, and its Protein Product P is a functional component of only Pathway P2 -> then Disease D directly perturbs Pathway P2<rdf:Description><log:is rdf:parseType=‘Quote’><rdf:Description rdf:about=‘variable#Gene_G’>

<hasProduct rdf:resource=‘variable#Protein_P’/><isImplicatedIn rdf:resource=‘variable#Disease_D’/>

</rdf:Description><rdf:Description rdf:about=‘variable#Protein_P’><inPathway rdf:resource=‘variable#Pathway_P2’/>

</rdf:Description><log:is><log:implies rdf:parseType=‘Quote’>

<rdf:Description rdf:about=‘variable#Disease_D’><D_perturbs rdf:resource=‘variable#pathway_P2’>

</rdf:Description></log:implies></rdf:Description>

Why Semantic Web for Life Sciences?

Heterogeneous data integration using explicit semanticsExpression well-defined and rich models of biological systemsAnnotating findings and interpretations formally and sharing with other scientists Embedding models and semantics within papersApplying logic to infer additional insights and to propose and/or capture new hypotheses

Q U E S T I O N SQ U E S T I O N SA N S W E R SA N S W E R S

RDF Support in Oracle RDBMS

Souripriya Das, Ph.D.Consultant Member of Technical Staff

Oracle New England Development Center

Overview

Three types of database objectsModel RDF graph consisting of a set of triplesRulebase Set of (user-defined) rulesRule Index Entailed RDF graph

We discuss following aspects for each type of objectDDLDMLViewsSecurity

RDF Query (with Inference)

RDF Models

Model: Overview

Each RDF Model (graph) consists of a set of triplesA triple (statement) consists of three components

– Subject URI or blank node– Predicate URI– Object URI or literal or blank node

A statement itself can be a resource (allowing nested graphs)

Model: Example

Family:(:John :brotherOf :Mary)(:John :age “16”^^xsd:Integer)(:Mary :parentOf :Matt)(:John :name “John”)(:Mary :name “Mary”)

Reification:(:John :thinks _:S1)(_:S1 rdf:subject :Sue)(_:S1 rdf:predicate :livesIn)(_:S1 rdf:object “NYC”)

:John

:Mary

brotherOf

:Matt

16

parentOf

age

:Sue NYClivesIn

thinks

RDF Query

SDO_RDF_MATCH Table FuncArguments

– Graph pattern A sequence of triple patternsTriple patterns typically use variables

– RDF Data set a set of models– Filter– Aliases

…FROM TABLE(SDO_RDF_MATCH(

‘(?x :brotherOf ?y) (?y :parentOf ?z)’,SDO_RDF_Models(‘family’),…

)) t…

SDO_RDF_MATCH: returnColumns (of type VARCHAR2) in each returned row:

For each variable ?x in Graph Pattern– x– x$rdfVTYP

URI, Literal, Blank node– x$rdfLTYP

Specific literal type (e.g., xsd:integer)– x$rdfCLOB

Contains actual value, if ?x matches a CLOB value

– x$rdfLANGLanguage tag, if any (e.g., “en-us”)

If no variable in Graph Pattern– A dummy column

SDO_RDF_MATCH: matchingMatching multiple representations

The same point in value space may have multiple representations

– “10”^^xsd:Integer– “10”^^xsd:PositiveInteger– “010”^^xsd:Integer– “000010”^^xsd:Integer

SDO_RDF_MATCH automatically resolves these

RDF Query: ExampleFind salary and hiredate of all the unclesSELECT emp.name, emp.salary, emp.hiredateFROM emp,

TABLE(SDO_RDF_MATCH(‘(?x :brotherOf ?y) (?y :parentOf ?z) (?x :name ?name)’,SDO_RDF_Models(‘family'),

…)) tWHERE emp.name=t.name;Use of SDO_RDF_MATCH allows embedding a graph query in a SQL query

RDF Query: Example 2Find pairs of persons residing at the same address where the first person rents a truck and the second person buys a fertilizerSELECT t3.x name1, t3.y name2FROM AddrTable t1, AddrTable t2,

TABLE(SDO_RDF_MATCH(‘(?x :rents ?a) (?a rdf:type :Truck)(?y :buys ?b) (?b rdf:type :Fertilizer)’,SDO_RDF_Models(‘Activities'),

…)) t3WHERE t1.name=t3.x and t2.name=t3.y and

t1.addr=t2.addr;

RDF Rulebases

Rulebase: Overview

Each RDF rulebase consists of a set of rulesEach rule consists of

– antecedent: graph-pattern– filter condition (optional)– Consequent: graph-pattern

One or more rulebases may be used with relevant RDF models (graphs) to obtain entailed graphs

Rulebase: Example

Rules in a rulebase family_rb:Antecedent: ‘(?x :brotherOf ?y) (?y :parentOf ?z)’Filter: NULLConsequent: ‘(?x :uncleOf ?z)’

Antecedent: ‘(?x :age ?a)’Filter: ‘a >= 65’Consequent: ‘(?x :ageGroup “Senior”)’

Antecedent: ‘(?x :parentOf ?y) (?y :parentOf ?z)’Filter: NULLConsequent: ‘(?x :grandParentOf ?z)’

RDF Rule Indexes

Rule Index: Overview

A rule index represents an entailed graphA rule index is created on an RDF dataset (consisting of a set of RDF models and a set of RDF rulebases)

Rule Index: Example

A rule index may be created on a dataset consisting of

– family RDF data, and– family_rb rulebase (shown earlier)

The rule index will contain inferred triples showing uncleOf and ageGroup information

RDF Query with Inference

SDO_RDF_MATCH with Rulebases

Arguments– Graph pattern

A sequence of triples (with variables)– RDF Data set

a set of modelsa set of rulebases

– Filter– Aliases

…FROM TABLE(SDO_RDF_MATCH(

‘(?x :uncleOf ?y)’,SDO_RDF_Models(‘family’),SDO_RDF_Rulebases (‘rdfs’, ‘family_rb’)…

)) t…

RDF Query w/ Inference: Example

Find salary and hiredate of all the unclesSELECT emp.name, emp.salary, emp.hiredateFROM emp,

TABLE(SDO_RDF_MATCH(‘(?x :uncleOf ?y) (?x :name ?name)’,SDO_RDF_Models(‘family'),SDO_RDF_Rulebases(‘rdfs’, ‘family_rb'),

…)) tWHERE emp.name=t.name;

RDF Query w/ Inference: Example 2

Find pairs of persons residing at the same address where the first person rents a truck and the second person buys a fertilizerSELECT t3.x name1, t3.y name2FROM AddrTable t1, AddrTable t2,

TABLE(SDO_RDF_MATCH(‘(?x :rents ?a) (?a rdf:type :Truck)(?y :buys ?b) (?b rdf:type :Fertilizer)’,SDO_RDF_Models(‘Activities'),SDO_RDF_Rulebases(‘rdfs’),

…)) t3WHERE t1.name=t3.x and t2.name=t3.y and

t1.addr=t2.addr;

RDF Models

Model: DDL

Procedures provided as part of the API may be used to

– Create a model– Drop a model

When a user creates a model, a database view gets created automatically

– rdfm_familyA model corresponds to a column of type SDO_RDF_TRIPLE_S in a base tableEach model has exactly one base table associated with it

Model: DDL Creating a Model

Create an Application TableCREATE TABLE family_table (

id NUMBER, family_triple SDO_RDF_TRIPLE_S);Create a Model

EXEC SDO_RDF.CREATE_RDF_MODEL(‘family’, ‘family_table’,‘family_triple’);Automatically creates the following database view

rdfm_family (…)

Loading RDF Data into Oracle

Java API provided to load NTriple into NDM

Sample XSLs provided– To convert RDF to NTriple– To convert RDF to INSERT statements

Model: DML

SQL DML commands may be used to do DML operations on a base table to effect DML (i.e., triple insert, delete, and update) on the corresponding model

Insert TriplesINSERT INTO family_table VALUES (1,

SDO_RDF_TRIPLE_S(‘family','<http://example.org/family/John>', '<http://example.org/family/brotherOf>', ‘<http://example.org/family/Mary>'));

Model: Security

The creator of the base table corresponding to a model can grant privileges to other usersTo perform DML to a model, a user must have DML privileges for the corresponding base tableThe creator of a model can grant QUERY privileges on the corresponding database view to other usersA user can query only those models for which s/he has QUERY privileges to the corr. database viewsOnly the creator of a model can drop the model

Model: Views

Database views corresponding to the models

RDF Rulebases

Rulebase: DDL

Procedures provided as part of the API may be used to

– Create a rulebasecreate_rulebase('family_rb');

– Drop a rulebase– drop_rulebase('family_rb');

When a user creates a rulebase, a database view gets created automatically

– rdfr_family_rb (rule_name, antecedent, filter, consequent, aliases)

Rulebase: DML

SQL DML commands may be used on the database view corresponding to a target rulebase to insert, delete, and update rulesinsert into mdsys.rdfr_family_rb values(‘uncle_rule',‘(?x :brotherOf ?y) (?y :parentOf ?z)’,NULL,'(?x :uncleOf ?z)', SDO_RDF_Aliases(…));

Rulebase: Security

Creator of a rulebase can grant privileges to the corresponding database view to other usersPerforming DML operations requires invoker to have appropriate privileges on the database viewOnly the creator of a rulebase can drop the rulebase

Rulebase: Views

RDF_RULEBASE_INFO– Contains the list of rulebases– For each rulebase, contains additional

information (such as, creator, view name, etc)Content of each rulebase is available from the corresponding database view

RDF Rule Indexes

Rule Index: DDL

Procedures provided as part of the API may be used to

– Create a rule indexcreate_rules_index ('family_rb_rix_family‘,

SDO_RDF_Models('family'),SDO_RDF_Rulebases(‘rdfs','family_rb'));

– Drop a rule indexdrop_rules_index ('family_rb_rix_family');

When a user creates a rule index, a database view gets created automatically

– rdfi_family_rb_rix_family (…)

Rule Index: Security

To create a rule index on an RDF dataset (models and rulebases), user needs to have QUERY privileges on those models and rulebasesCreator of a rule index holds QUERY privilege on the rule index and may grant this privilege to other usersOnly the creator of a rule index can drop it

Rule Index: Views

RDF_RULEINDEX_INFO– Contains the list of rule indexes– For each rule index, contains additional

information (such as, creator, status, etc)RDF_RULEINDEX_DATASETS

– For every rule index, stores the names of its models and rulebases

Rule Index: Dependencies

Content of a rule index depends upon the content of each element of its dataset

– Any modification to the models or rulebases in its dataset invalidates the rule index

– Dropping a model or rulebase will drop dependent rule indexes automatically.

Summary

RDF Data Model– Models (Graphs)– RDF Query using SDO_RDF_MATCH Table Function

RDF Data Model with (user-defined) Rules– Models (Graphs)– Rulebases– Rule Indexes– RDF Query on entailed RDF graphs

Management (DDL, DML, Security, …)– Models, Rulebases, and Rule Indexes

RDF Data Model Demo

Demo: Family Schema

Demo: Family Schema 2

Demo: Family Model Data

Demo: Family Model Data (Alt)

Demo: Query without Inferenceselect m from TABLE(SDO_RDF_MATCH(

'(?m rdf:type :Male)',SDO_RDF_Models('family'),null,SDO_RDF_Aliases(SDO_RDF_Alias('', 'http://www.example.org/family/')),

null));M--------------------------------------------------------------------------------http://www.example.org/family/Jackhttp://www.example.org/family/Tom

Demo: Query w/ RDFS Inference select m from TABLE(SDO_RDF_MATCH(

'(?m rdf:type :Male)',SDO_RDF_Models('family'),SDO_RDF_Rulebases(‘RDFS’),SDO_RDF_Aliases(SDO_RDF_Alias('', 'http://www.example.org/family/')),

null));M--------------------------------------------------------------------------------http://www.example.org/family/Jackhttp://www.example.org/family/Tomhttp://www.example.org/family/Johnhttp://www.example.org/family/Matthttp://www.example.org/family/Sammy

Demo: Family RulebaseAntecedent: ‘(?x :parentOf ?y) (?y :parentOf ?z)’Filter: NULLConsequent: ‘(?x :grandParentOf ?z)’

Demo: Query w/ Family and RDFS Inference

select x, y from TABLE(SDO_RDF_MATCH('(?x :grandParentOf ?y) (?x rdf:type :Male)',SDO_RDF_Models('family'),SDO_RDF_Rulebases('RDFS','family_rb'), SDO_RDF_Aliases(SDO_RDF_Alias('','http://www.example.org/family/')),

null));X Y------------------------------------------------------ -----------------------------------------------------http://www.example.org/family/John http://www.example.org/family/Cindyhttp://www.example.org/family/John http://www.example.org/family/Tomhttp://www.example.org/family/John http://www.example.org/family/Jackhttp://www.example.org/family/John http://www.example.org/family/Cathy

Q U E S T I O N SQ U E S T I O N SA N S W E R SA N S W E R S

Demo of Siderean’s Seamark Navigation Server

Mike DiLascio & Joanne Luciano

Agenda

About Siderean Software & Predictive Medicine, Inc.Introducing Seamark Navigation Server v.3.6Seamark & Oracle 10g RDF Data Model Demonstration of Seamark / Oracle 10g integrationLessons Learned / Q&A

About Siderean Software

Aggregate, organize and navigate information -the way users think –

-to improve analysis and decision making.

Founded in 2001 and based in El Segundo, CAVentured backed in 2004Delivering RDF-centric navigation and analysis capabilities for end users (a.k.a. - “the last mile”)Active W3C member leveraging Semantic Web standardsDemonstrating integrated Seamark navigation layer over Oracle 10g RDF Data Model in collaboration with Predictive Medicine, Inc.

Current solutions“50,000 results!!! Now what?” “I give up! Hello? Get me an apple!” “Why do I get oranges when I’m looking

for apples?”

CONTENT PRODUCER:“I just produced three apples

last week!”

Knowledge management –breathtakingly expensive

Enterprise search –a brute force approach

IT:“As soon as I fix his, hers stops working.”

Introducing Seamark Navigation Server“I can see the big picture!” “No more staring at a blank text box.” “I can drill down quickly to what I want.”

CONTENT PRODUCER:“I knew we had an apple in

here somewhere.”

Seamark – layering organization to deliver pinpoint navigation

IT:“I can take my coffee

break now.”

How it works: process

Term

Event

Person

PlaceText

View View

Metadata about data and content is aggregated…

Organized into a unified information architecture…

Analyzed to generate on-demand views…

Providing pinpoint navigation across

the data and content

How it works: architectureUser Navigation

and User Tagging

Unstructured Content and Data Feeds

Search Engines

Structured Content Sources

MetadataAggregator

NavigationMetadata

NavigationWeb Services

Web Browsers& Portals

User Alerts

Feed Aggregators

Seamark/Oracle integration architecture: Phase 1

User Navigationand User Tagging

Web Browsers& Portals

Feed Aggregators

CachedNavigationMetadata

NavigationWeb Services

User Alerts

Oracle 10g RDF Data Model for scalable

persistence of metadata

Batch RDFMatchQuery issued from

Seamark at index time

Seamark/Oracle integration architecture: Phase 2

User Navigationand User Tagging

DynamicNavigation Metadata

NavigationWeb Services

Web Browsers& Portals

Federated RDFMatchQueries issued from

Seamark at query time

User Alerts

Oracle 10g RDF Data Model for scalable

persistence of metadata Feed Aggregators

Seamark Demo: Background & ConceptsLife Sciences demonstration premise

RDF offers high value during early stage research

Leveraging strengths of Oracle 10g & Seamark v3.6Oracle – large datasets / scalabilitySeamark – useful subsets / flexible navigation & insights

Project elapsed time - about one week Locating and identifying data sources represented the

greatest time elementData sources in RDF required minimal integration timeNon-RDF data sources required transformation and linking

values (non-trivial but straightforward)

Seamark Demonstration: Identification of new drug candidates

GO2Keyword.rdf

UniProt.rdf

GO.rdf

Keywords.rdf

Taxonomy.rdfPubMed.xml

Citation

IntAct.rdf

Organism

Enzymes.rdf

OMIM.rdf

GO2OMIM.rdf

GO2Enzyme.rdf

MIM Id

KEGG.rdf

KeywordGO2UniProt.rdf

Protein

Enzyme

ProbeSet.rdf

Gene

Probe

Pathway

Compound

1. Differentiate different forms of disease2. Identify patients subgroups. 3. Identify top biomarkers4. Identify function5. Identify biological and chemical properties and disease associations of biomarker 6. Identify documents7. Identify role in metabolic pathways8. Identify compounds that interact9. Identify and compare function in other organisms10. Identify any prior art

Live Seamark Life Sciences Demonstration:

Sample Screenshots

Seamark application start page shows integration of OMIM, GO, KEGG, UniProt and NCBI

Select: Probe Set ID: “M18255_cds2_s_at”

Cytoplasm 1st of 9 MatchesCellular Location Via Gene Ontology

Results: 9 Matches on “M18255_cds2_s_at” to the Gene Ontology

Page Scroll

Cytoplasm 1st of 9 Matches

Page Scroll

Plasma Membrane, …, 2nd of 9 MatchesCellular Location Via Gene Ontology

Page Scroll for more results, etc.

Cytoplasm 1st of 9 Matches

Start Page: Optionally search across entire collection based upon keywords from the integrated data sources

Seamark Lessons LearnedRDF offers multiple unconstrained views of data/relationships– Provides maximum flexibility during early stage research – Later stages can leverage OWL to constrain known

relationships

Data providers – Timing is right to publish in RDF format– Cut your customer’s integration costs– Speed discovery time

Even with one week of effort…– Proof of Concept demonstrates value of broad & deep

integration – Additional value in extending POC in customer pilot initiatives

Siderean Seamark Conclusion

Getting the precise information we need from today’s data glut is profoundly difficultSolving this problem requires a solution that works the way you thinkSiderean is the world’s first turnkey navigation server for the enterprise and people at large

To arrange a demonstration of Seamark or for more information please contact:

Mike DiLascioOffice: +1 781 652 0339Mobile: +1 781 354 7663mdilascio@siderean.com

Thank You!

Siderean Software, Inc.390 North Sepulveda Blvd., Suite 2070El Segundo, CA 90245-4475 USAhttp://www.siderean.com