Relational Patterns in OWL and their application to OBO

Relational Patterns in OWL and their application to OBO

Robert Hoehndorf1, Anika Oellrich1, Michel Dumontier2,

Janet Kelso4, Heinrich Herre3, and Dietrich Rebholz-Schuhmann1

OWLED2010:San Francisco:Relational Patterns in OWL and their application to OBO 1

1European Bioinformatics Institute, 2Carleton University, 3University of Leipzig, 4Max Planck Institute for Evolutionary Anthropology

Ontology

• Explicit specification of conceptualization – specify meaning of terms in vocabulary

• formal ontology uses languages with explicit semantics to specify meaning– axioms in FOL, OWL, etc


Biomedical Ontologies

• OBO - 100+ ontologies • OBO Foundry

– common principles and criteria– format: OBOF or OWL – must use OBO Relation Ontology


OBO Flatfile Format

Graph based language• Nodes [term] represent categories

[Term]id: CL:0000028name: CNS neuron (sensu Nematoda and Protostomia)is_a: CL:0000540 ! neuronrelationship: develops_from CL:0000338


OBOF examples

• Edges [typedef] represent relations

[Typedef] id: develops_fromis_transitive: true

OBOF semantics not explicit• One solution uses tailored FOL semantics• Another uses fixed semantics for relations and relational

statements


OBO Relationship Ontology

• Smith, 2005 • FOL definitions for common relations

– C part_of C' if and only if: given any c that instantiates C at a time t, there is some c' such that c' instantiates C' at time t, and c *part_of* c' at t.

– weak axioms (reflexivity, transitivity, etc.) – use of FOL and ternary relations: mapping to

(decidable) OWL problematic


Horrocks OBO2OWL semantics

• Fixed semantics via mapping to OWL– intersection, union, disjointness

• R(C,D) maps to C subClassOf R some D

• OBOF[Term]name: mouserelationship: has-part tail

• OWLmouse subClassOf has-part some tail


Critique

• Horrocks semantics does not use RO semantics – but RO must be used for OBO Foundry ontologies

• Horrocks semantics sometimes inadequate

Tailless-Mouse subClassOf lacks-part some Tail

Should instead be:

Tailless-Mouse subClassOf not(has-part some Tail)


Problem statement

• RO in FOL -> undecidable • RO not integrated in OBOF

– no semantics for OBOF• static translation of relations to OWL

C subClassOf R some D -> inadequate– develop OBOF semantics, preferably using

decidable logic (OWL)


Proposal

• Observation: relation between classes defined through relations between instances

R(C,D) <-> Def(C,D) • C and D occur as variables in Def • Idea: Pattern Definitions in OWL

– extend OWL with variables for classes– use Manchester OWL syntax– formulate relations using extended OWL syntax – similar to OWL Pre-Processing Language (OPPL)


Extended OBOF syntax:OWLDEF

[Typedef]id: CC-has-partname: has-partowldef: ?X subClassOf: has-part some ?Y• every use of the relation CC-has-part in the OBOF is expanded to

an OWL axiom in which the variables are filled by the classes between which the relation was asserted.

[Term]id: Mouserelationship: has-part Tailgives:

Mouse subClassOf has-part some Tail


integral-part-of

integral-part-of(?X, ?Y): ?X part-of ?Y and ?Y has-part ?X

• Produces two OWL class axioms?X subClassOf part-of some ?Y ?Y subClassOf has-part some ?X

• Which are transformed to one(?X and not (part-of some ?Y)) or(?Y and not (has-part some ?X)) subClassOf Nothing



OBO2OWL

• prototype extends OWL API to convert OBOF to OWL with owldef statements

• ?X and ?Y from the owldef statement are replaced by the corresponding term names

• inline axiom parser generates an OWL axiom• API and web-interface at

http://bioonto.de/obo2owl


http://bioonto.de/obo2owl

OWL2OBO

• Extract relational patterns from OWL ontology – Naïve: replace ?X and ?Y with all combinations of

named classes are generated to fill variables in the relation patterns and generate OWL axioms from binary class relations

– Hermit used to prove axioms; keep track of those which could be inferred from asserted axioms

– Obtain list of theorems that can be added to OBO file


Evaluation

• Celltype Ontology– 1253 is-a; 275 develops-from relations– identified 9,497 is-a and 124,420 develops-from

• Malaria Ontology– 3 realized-by– Identified 56 realized-by

• Sequence Ontology– Added 1 exon integral-part-of transcript– Now: all exons are part-of transcript and all transcripts

has-part exon


Results @ http://bioonto.de/obo2owl

Open Problems

• intersection, union statements and relations have unclear semantics

[Term]id: ID:Aintersection_of: ID:Bintersection_of: integral-part-of ID:C

• OWL class axiom (integral-part-of) intersected with class description -> invalid

• (A subClassOf B) intersected with C: is C intersected with A, B, both?

• no obvious solution: inherent problem in OBOF, needs to be addressed (by OBOF developers)


Advanced Applications

• n-ary relations between classes• introduce variables ?X1, ..., ?XN• example:

has-quality-with-value(?X1, ?X2, ?X3): ?X1 subClassOf has-quality some

(?X2 and has-value some ?X3)


Advanced Applications 2

• open patterns as meta-properties of ontologies• Consider the combinatorics of connecting

chemical groups to base groups for high throughput chemical synthesis

• assert ?X subClassOf is-connected-to some ?Y• semantics: all named classes in ontology satisfy

the pattern: all classes can be connected to all other classes


Future Research

• integration with OPPL2 and use of OPPL2 to define relations

• formalization of further relations• repository of frequently used relation

definitions• efficient conversion from OWL to OBO• integration with OWL: use of annotation

properties


Conclusion

• pattern approach as a simple interface to complex definitions

• flexible semantics based on conversion to OWL– we implemented OBO RO in OWL

• current approaches have shortcomings – open issues with intersection, union, etc

• http://bioonto.de/obo2owl


Technology

Relational Patterns in OWL and their application to OBO