jena_reasoner

Try to understand how ontology integration works in Apache Jena.

Two possibilities are investigated in this project:

• via Apache Jena Fuseki, a SPARQL server
• via Apache Jena API, a Java API

1. Data

Two samples with corresponding ontology data are tested:

	Sample data	Ontology
Pizza	data/pizza.ttl	data/pizza_ontology.rdf
Biology	data/biology.ttl	data/biolink-model.owl.ttl

1.1. Pizza

We focus on the inverse relationships defiend in OWL:

:isIngredientOf owl:inverseOf :hasIngredient .

i.e. the property isIngredientOf is the inverse of the property hasIngredient (see data/pizza_ontology.rdf).

Two pizza Margherita and Cheese Bacon: the relationship between a pizza and an ingredient is declared by the property either isIngredientOf or hasIngredient (see data/pizza.ttl). In summary:

• Margherita contains tomato and basilic
• Cheese Bacon contains cheese, bacon, and tomato

1.2. Biology

Biolink Model is a high level data model of biological entities and their associations in SWRL: a Semantic Web Rule Language Combining OWL and RuleML.

We focus on the inverse relationships:

biolink:catalyzes owl:inverseOf biolink:has_catalyst .

i.e. the property biolink:catalyzes is the inverse of the property biolink:has_catalyst (see data/biolink-model.owl.ttl).

Three biological reactions have corresponding catalysts (enzymes) and their relationships are declared by the property either biolink:catalyzes or biolink:has_catalyst (see data/biology.ttl). In summary:

• ReactionA is catalyzed by EnzymeA1
• ReactionB is catalyzed by EnzymeB1 and EnzymeB2
• ReactionC is catalyzed by EnzymeC1 and EnzymeC2

2. Expected result

Initial missing information is complemented by ontology transitivity.

e.g. for Pizza, after a SPARQL query asking all triples ?pizza pizza:hasIngredient ?ingredient, initial missing information will be complemented because ex:tomato pizza:isIngredientOf ex:margherita is equivalent to ex:margherita pizza:hasIngredient ex:tomato.

3. Method

3.1. Fuseki

Each type of data (Pizza, Biology) is uploaded on TDB, a component of Jena for semantic data storage and query.
e.g. pizza_ontology.rdf is saved in /opt/tomcat/fuseki/base/databases/test_ontology_pizza.

Then a Fuseki's configuration is done according to Fuseki: Configuring Fuseki (section Inference) to integrate ontology. The configuration is the same as data/config.ttl.

All possible reasoners are tested:

• Generic Rule Reasoner: http://jena.hpl.hp.com/2003/GenericRuleReasoner
• Transitive Reasoner: http://jena.hpl.hp.com/2003/TransitiveReasoner
• RDFS Rule Reasoner: http://jena.hpl.hp.com/2003/RDFSExptRuleReasoner
• Full OWL Reasoner: http://jena.hpl.hp.com/2003/OWLFBRuleReasoner
• Mini OWL Reasoner: http://jena.hpl.hp.com/2003/OWLMiniFBRuleReasoner
• Micro OWL Reasoner: http://jena.hpl.hp.com/2003/OWLMicroFBRuleReasoner

Note: there is no reasoner for SWRL.

3.2. Jena API

Java codes are in src/Pizza.java and src/Biology.java.
For both case Jena used Full OWL Reasoner (OWLFBRuleReasoner).

4. Results

4.1. Fuseki

No ontology integration. Only raw information is retrieved:

# Pizza
[http://example.com/cheese_bacon, http://example.com/pizzas.owl#hasIngredient, http://example.com/tomato]
[http://example.com/cheese_bacon, http://example.com/pizzas.owl#hasIngredient, http://example.com/bacon]
[http://example.com/margherita, http://example.com/pizzas.owl#hasIngredient, http://example.com/basilic]

4.2. Jena API

The expected results are achieved.

--- In raw input pizza data ---
hasIngredient property
[http://example.com/cheese_bacon, http://example.com/pizzas.owl#hasIngredient, http://example.com/tomato]
[http://example.com/cheese_bacon, http://example.com/pizzas.owl#hasIngredient, http://example.com/bacon]
[http://example.com/margherita, http://example.com/pizzas.owl#hasIngredient, http://example.com/basilic]
isIngredientOf property
[http://example.com/cheese, http://example.com/pizzas.owl#isIngredientOf, http://example.com/cheese_bacon]
[http://example.com/tomato, http://example.com/pizzas.owl#isIngredientOf, http://example.com/margherita]

--- Results of SPARQL on raw model ---
( ?s = <http://example.com/cheese_bacon> ) ( ?o = <http://example.com/bacon> )
( ?s = <http://example.com/cheese_bacon> ) ( ?o = <http://example.com/tomato> )
( ?s = <http://example.com/margherita> ) ( ?o = <http://example.com/basilic> )

Binding ontology schema
    reasoner: org.apache.jena.reasoner.rulesys.OWLFBRuleReasoner@65004ff6
--- Results of SPARQL on inferred model ---
( ?s = <http://example.com/cheese_bacon> ) ( ?o = <http://example.com/bacon> )
( ?s = <http://example.com/cheese_bacon> ) ( ?o = <http://example.com/cheese> )
( ?s = <http://example.com/cheese_bacon> ) ( ?o = <http://example.com/tomato> )
( ?s = <http://example.com/margherita> ) ( ?o = <http://example.com/basilic> )
( ?s = <http://example.com/margherita> ) ( ?o = <http://example.com/tomato> )

--- In raw input biology data ---
has_catalyst property
[http://example.com/reactionB, https://w3id.org/biolink/vocab/has_catalyst, http://example.com/enzymeB1]
[http://example.com/reactionA, https://w3id.org/biolink/vocab/has_catalyst, http://example.com/enzymeA1]
catalyzes property
[http://example.com/enzymeC2, https://w3id.org/biolink/vocab/catalyzes, http://example.com/reactionC]
[http://example.com/enzymeC1, https://w3id.org/biolink/vocab/catalyzes, http://example.com/reactionC]
[http://example.com/enzymeB2, https://w3id.org/biolink/vocab/catalyzes, http://example.com/reactionB]

--- Results of SPARQL on raw model ---
( ?s = <http://example.com/reactionA> ) ( ?o = <http://example.com/enzymeA1> )
( ?s = <http://example.com/reactionB> ) ( ?o = <http://example.com/enzymeB1> )

Binding ontology schema
    reasoner: org.apache.jena.reasoner.rulesys.OWLFBRuleReasoner@1ddd3478
    duration: 00:00:59
Creating inference model
    duration: 00:00:00
--- Results of SPARQL on inferred model ---
( ?s = <http://example.com/reactionA> ) ( ?o = <http://example.com/enzymeA1> )
( ?s = <http://example.com/reactionB> ) ( ?o = <http://example.com/enzymeB1> )
( ?s = <http://example.com/reactionB> ) ( ?o = <http://example.com/enzymeB2> )
( ?s = <http://example.com/reactionC> ) ( ?o = <http://example.com/enzymeC1> )
( ?s = <http://example.com/reactionC> ) ( ?o = <http://example.com/enzymeC2> )
    duration: 00:00:00

5. Conclusion

• The ontology inference was not successful on Fuseki server.
  • The configuration file is not correctly applied.
  • I need advice from the experts...!!!
• Inference was successful via Jena API.
  • For a big ontology data (Biolink Model), the step of bindSchema() is slow (around 1 minute)
  • Full OWL Reasoner seems working on Biolink Model (SWRL format).

Future work

• Merge/bind multiple ontology files to get a one ontology model
• Save ontology model (then read it when needed)
• Investigate whether Full OWL Reasoner reasoner application loses information of SWRL format