Reasoning the FMA Ontologies with TrOWL - PowerPoint Presentation

Slide1

Reasoning the FMA Ontologies with TrOWL

Jeff Z. Pan,

Yuan Ren

, Nophadol Jekjantuk, and Jhonatan Garcia

University of Aberdeen, UK

ORE2013Slide2

The FMA ontology

The Foundational Model of Anatomy ontology is “an evolving computer-based knowledge source for biomedical informatics”

Developed with Protégé as a FRAME-BASED system

Consists of several components such as MetaknowledgeEvolves (latest version released in 2010)Highly expressiveSeveral OWL translationsDLR and FullR: OWL DL/FULL versions without/with metamodellingConstitutional: alternative OWL DL translation with metamodellingOWL2G_noMTC: OWL2 translation from FAM 3.0 without metamodellingDLR_M1/M2: portion of DLR enriched with the class-based approach (Glimm et al., 2010) to accommodate metaclassesSlide3

TrOWL: Tractable reasoning infrastructure for OWL 2

Semantic Approximation (AAAI2007)

Pre-compute and compile the materialisation of OWL 2 ontologies in OWL 2 QL

Sound and complete for conjunctive queries without non-distinguished variablesTractable in run-timeSyntactic Approximation (AAAI2010)Normalise OWL 2 axioms into nominal-safe EL++ with additional data structures to maintain non-EL semanticsApproximate deduction on the normalisation resultsSound, incomplete but practically high recall for many ontologiesTractable TBox classification and ABox materialisationOracle 11g support, SPARQL 1.1 query answering (leveraging OWL-BGP), local closed world reasoning, Jena API, etc.Slide4

Syntactic Approximation

Normalisation

Representing non-EL expressions with fresh names

Maintain complementary relationsDeductionCEL rulesAdditional rulesE.g. A subClassOf B => not B subClassOf not AExample ontology:A subClassOf forall r Bforall r C subClassOf DB subClassOf

C

=>

A

subClassOf

D

ALL

r

B

A

C

ALL

D

Some

r

nB

A

nC

Some

D

B

C

X1

X2Slide5

Metamodelling in FMA Ontology

FMA frame-based ontology contains

metamodelling

E.g. Physical_anatomical_entity instanceOf Anatomical_entity_templatePhysical_anatomical_entity subClassOf Anatomical_entityDifferent implementations in OWL ontologiesFMA FullR uses OWL Full;FMA Consititutional encodes metaclass assertions with class subsumptions, metaproperty

assertions with existential and universal restrictions;

OWL 2 DL with punning semantics

A class and an individual with same IRI will still be treated as different entities, leading to incomplete results

OWL 2 DL with class-based approach

Introducing representative individual of each concept

Encoding

subsumptions/class assertions with object property relationsSlide6

Evaluation Results

FMA ontologies are in general very difficult to reason with

Especially with

Metamodelling involvedTrOWL performs generally well on FMA ontologiesGenerally faster than fully-fledged, universal, intractable reasoners;The only one to classify FMA-OWL2G_noMTC TBox in 1 hour;Practically high recallSlide7

Dealing with

Unsatisfiable

Concepts

Translated versions of FMA contain many unsatisfiabilitiesFMA Constitutional: 33,433 / 41,648FMA OWL2G_noMTC: 67,771 / 85,005Investigating such unsatisfiabilities is difficultHard to compute justificationsRequires a lot of entailment checkingsToo many unsatisfiability to look intoWe want to get into the core of the problem efficientlySlide8

Just. (A

subClassOf

Bot)Finding the Core UnsatisfiabilitiesKalyanpur et al.’s root and derived unsatisfiable conceptsB is parent of AA is derivedNon-derived unsatisfiable

concept is root

A derived concept can have alternative justification that contains no parent

Eliminating all root concepts do not necessarily eliminate all

unsatisfiability

Still need to compute justifications and entailment

checkings

Just. (B

subClassOf Bot)Slide9

Finding the Core Unsatisfiabilities

Type I and Type II

unsatisfiable

conceptsPurely based on the derivation relations between axiomsSuitable with a forward-chaining completion-based algorithmType I concepts are full-unsatisfiable in reasoningType II concepts are semi-unsatisfiable in reasoningnot immediately subsumed by all conceptspropagates Type IICan become Type I if appropriate inference occurs

axiom1

axiom2

axiom3

A

subClassOf

Bot

B

subClassOf

Bot

Type I

Type II

Type I andSlide10

Application on FMAs

Repairing the Type I concepts will resolve all existing

unsatisfiabilities

From TrOWL’s perspectiveFewer enough Type I makes debugging much easierE.g. 145 Type I in FMA Constitutional, only 0.43% of all the unsatisfiable concepts; 6 axioms directly involved, out of the 122,136 logical axiomsSlide11

Summary and Future Work

TrOWL and its syntactic approximation facility is well suited for the reasoning,

metamodelling

and debugging of the FMA ontologiesStriking a balance among expressiveness, performance and qualityFuture worksA completeness-guarantee?Why does TrOWL have high recalls on certain ontologies?A potential tractable DL that covers FMA family?A fully-fledged completion-based reasoner for OWL2 DL?Will be intractableParallelisation?Changing CEL rules to ELK rules?Parallelising the additional approximate deduction rulesImproved entailment checkingCurrently using the dual-ontology classification algorithm from CELChanging to a goal-driven algorithm?Slide12

Thank You!

http

://trowl.eu