Ontology-Based Classification of Molecules: a Logic Programming Approach

ONTOLOGY-BASED

CLASSIFICATION OF MOLECULES:A LOGIC PROGRAMMING APPROACH

Despoina Magka

Department of Computer Science, University of Oxford

November 30, 2012

BIOINFORMATICS AND SEMANTIC TECHNOLOGIES

Life sciences data deluge

Hierarchical organisation of biochemical knowledge

Fast, automatic and repeatable classification driven bySemantic technologiesWeb Ontology Language, a W3C standard familyof logic-based formalismsOWL bio- and chemo-ontologies widely adopted

1


Life sciences data delugeHierarchical organisation of biochemical knowledge


1



Fast, automatic and repeatable classification driven bySemantic technologies

Web Ontology Language, a W3C standard familyof logic-based formalismsOWL bio- and chemo-ontologies widely adopted

1



Fast, automatic and repeatable classification driven bySemantic technologiesWeb Ontology Language, a W3C standard familyof logic-based formalisms

OWL bio- and chemo-ontologies widely adopted

1




1

THE CHEBI ONTOLOGY

OWL ontology Chemical Entities of Biological Interest

Dictionary of molecules with taxonomical informationPharmaceutical design and study of biological pathways

ChEBI is manually incremented

Currently ~30,000 chemical entities, expands at 3,500/yr

Existing chemical databases describe millions of molecules

Speed up growth by automating chemical classification

2

THE CHEBI ONTOLOGY

OWL ontology Chemical Entities of Biological InterestDictionary of molecules with taxonomical information

Pharmaceutical design and study of biological pathways





2

THE CHEBI ONTOLOGY


caffeine is a cyclic molecule


ChEBI is manually incrementedCurrently ~30,000 chemical entities, expands at 3,500/yrExisting chemical databases describe millions of moleculesSpeed up growth by automating chemical classification

2

THE CHEBI ONTOLOGY


serotonin is an organic molecule



2

THE CHEBI ONTOLOGY


ascorbic acid is a carboxylic ester



2

THE CHEBI ONTOLOGY

OWL ontology Chemical Entities of Biological InterestDictionary of molecules with taxonomical informationPharmaceutical design and study of biological pathways





2

EXPRESSIVITY LIMITATIONS OF OWL

1 At least one tree-shaped model for each consistent OWLontology problematic representation of cycles

2 No minimality condition on the models hard to axiomatiseclasses based on the absence of attributes

EXAMPLE

Cyclobutane v ∃(= 4)hasAtom.(Carbon u ∃(= 2)hasBond.Carbon)

C C

CC

Oxygen

OWL-based reasoning support

1 Is cyclobutane a cyclic molecule?2 Is cyclobutane a hydrocarbon?

Required reasoning support


3




EXAMPLE


C C

CC

Oxygen

OWL-based reasoning support1 Is cyclobutane a cyclic molecule? 8

2 Is cyclobutane a hydrocarbon?



3




EXAMPLE


C C

CC

Oxygen

OWL-based reasoning support1 Is cyclobutane a cyclic molecule? 82 Is cyclobutane a hydrocarbon? 8



3




EXAMPLE


C C

CC

Oxygen





3




EXAMPLE


C C

CC

Oxygen



Required reasoning support1 Is cyclobutane a cyclic molecule?2 Is cyclobutane a hydrocarbon?

3




EXAMPLE


C C

CC

Oxygen



Required reasoning support1 Is cyclobutane a cyclic molecule? 32 Is cyclobutane a hydrocarbon? 3

3

RESULTS OVERVIEW

1 Expressive and decidable formalism for modelling complexobjects: Description Graphs Logic Programs

2 Modelling that spans a wide range of structure-dependentclasses of molecules

3 Implementation that draws upon DLV and performsstructure-based classification with a significant speedup

4 Evaluation over part of the manually curated ChEBIontology revealed modelling errors

Language for representing biochemical structures with afavourable performance/expressivity trade-off

4

CLASSIFYING STRUCTURED OBJECTS

hasAtomsingledouble

0ascorbicAcid :

1

o

4 o 3 o

7

c2

o

8

c9

c

5

o

12

c 11

c6

o

10

c

13

h

Input fact: ascorbicAcid(a)Stable model: ascorbicAcid(a), hasAtom(a,af

i) for 1 ≤ i ≤ 13,o(af

i) for 1 ≤ i ≤ 6, c(afi) for 7 ≤ i ≤ 12, h(af

13), single(af8,a

f3),

single(af9,a

f4), single(af

12,afi) for i ∈ {5, 11}, single(af

11,af6),

single(af10,a

fi) for i ∈ {1, 9, 11, 13}, single(af

7,afi) for i ∈ {1, 8},

double(af2,a

f7), double(af

8,af9), horc(af

i) for 7 ≤ i ≤ 13,polyatomicEntity(a), carboxylicEster(a), cyclic(a)

Ascorbic acid is a cyclic polyatomic entity and a carboxylic ester

5


hasAtomsingledouble

0ascorbicAcid :

1

o

4 o 3 o

7

c2

o

8

c9

c

5

o

12

c 11

c6

o

10

c

13

h

ascorbicAcid(x)→hasAtom(x, f1(x)) ∧ . . . ∧ hasAtom(x, f13(x))o(f1(x)) ∧ . . . ∧ c(f7(x)) ∧ . . .∧single(f1(x), f7(x)) ∧ double(f7(x), f2(x)) ∧ . . .


i) for 1 ≤ i ≤ 13,o(af


13), single(af8,a

f3),

single(af9,a

f4), single(af


11,af6),

single(af10,a

fi) for i ∈ {1, 9, 11, 13}, single(af

7,afi) for i ∈ {1, 8},

double(af2,a

f7), double(af

8,af9), horc(af



5


hasAtomsingledouble

0ascorbicAcid :

1

o

4 o 3 o

7

c2

o

8

c9

c

5

o

12

c 11

c6

o

10

c

13

h

ascorbicAcid(x)→hasAtom(x, f1(x)) ∧ . . . ∧ hasAtom(x, f13(x))o(f1(x)) ∧ . . . ∧ c(f7(x)) ∧ . . .∧single(f1(x), f7(x)) ∧ double(f7(x), f2(x)) ∧ . . .

hasAtom(x, y1) ∧ hasAtom(x, y2) ∧ y1 6= y2 → polyatomicEntity(x)

∧5i=1hasAtom(x, yi) ∧ c(y1) ∧ o(y2) ∧ o(y3)∧

c(y4) ∧ horc(y5) ∧ double(y1, y2)∧single(y1, y3) ∧ single(y3, y4) ∧ single(y1, y5)→ carboxylicEster(x)


i) for 1 ≤ i ≤ 13,o(af


13), single(af8,a

f3),

single(af9,a

f4), single(af


11,af6),

single(af10,a

fi) for i ∈ {1, 9, 11, 13}, single(af

7,afi) for i ∈ {1, 8},

double(af2,a

f7), double(af

8,af9), horc(af



5


hasAtomsingledouble

0ascorbicAcid :

1

o

4 o 3 o

7

c2

o

8

c9

c

5

o

12

c 11

c6

o

10

c

13

h


i) for 1 ≤ i ≤ 13,o(af


13), single(af8,a

f3),

single(af9,a

f4), single(af


11,af6),

single(af10,a

fi) for i ∈ {1, 9, 11, 13}, single(af

7,afi) for i ∈ {1, 8},

double(af2,a

f7), double(af

8,af9), horc(af



5

CHEMICAL CLASSES WE COVERED

1 Existence of subcomponents

Carbon moleculesCarboxylic acids and carboxylic estersKetones and aldehydes

2 Exact cardinality of partsExactly two carbonsDicarboxylic acid

3 Exclusive compositionInorganic moleculesHydrocarbonsSaturated molecules

4 Cyclicity-related classesBenzenesCyclic moleculesAlkanes

6


1 Existence of subcomponentsCarbon molecules

Carboxylic acids and carboxylic estersKetones and aldehydes




6


1 Existence of subcomponentsCarbon moleculesCarboxylic acids and carboxylic esters

Ketones and aldehydes




6


1 Existence of subcomponentsCarbon moleculesCarboxylic acids and carboxylic estersKetones and aldehydes




6



2 Exact cardinality of parts

Exactly two carbonsDicarboxylic acid



6



2 Exact cardinality of partsExactly two carbons

Dicarboxylic acid



6






6




3 Exclusive composition

Inorganic moleculesHydrocarbonsSaturated molecules


6




3 Exclusive compositionInorganic molecules

HydrocarbonsSaturated molecules


6




3 Exclusive compositionInorganic moleculesHydrocarbons

Saturated molecules


6






6





4 Cyclicity-related classes

BenzenesCyclic moleculesAlkanes

6





4 Cyclicity-related classesBenzenes

Cyclic moleculesAlkanes

6





4 Cyclicity-related classesBenzenesCyclic molecules

Alkanes

6






6

EMPIRICAL EVALUATION

Draws upon DLV, a deductive databases engine

Evaluation with data extracted from ChEBI500 molecules under 51 chemical classes in 40 secsQuicker than other approaches:

[Hastings et al., 2010] 140 molecules in 4 hours[Magka et al., 2012] 70 molecules in 450 secs

Subsumptions exposed by our prototype:

ascorbic acid is a polyatomic entity, a carboxylic ester and acyclic moleculemissing from the ChEBI OWL ontology

Contradictory subclass relation from ChEBI:

Ascorbic acid is asserted to be a carboxylic acid (release 95)Not listed among the subsumptions derived by our prototype

7


Draws upon DLV, a deductive databases engineEvaluation with data extracted from ChEBI

500 molecules under 51 chemical classes in 40 secsQuicker than other approaches:






7


Draws upon DLV, a deductive databases engineEvaluation with data extracted from ChEBI500 molecules under 51 chemical classes in 40 secs

Quicker than other approaches:






7


Draws upon DLV, a deductive databases engineEvaluation with data extracted from ChEBI500 molecules under 51 chemical classes in 40 secsQuicker than other approaches:






7




Subsumptions exposed by our prototype:ascorbic acid is a polyatomic entity, a carboxylic ester and acyclic moleculemissing from the ChEBI OWL ontology



7




Subsumptions exposed by our prototype:ascorbic acid is a polyatomic entity, a carboxylic ester and acyclic moleculemissing from the ChEBI OWL ontology

Contradictory subclass relation from ChEBI:Ascorbic acid is asserted to be a carboxylic acid (release 95)Not listed among the subsumptions derived by our prototype

7

CONCLUSION AND FURTHER RESEARCH

Results1 Expressive and decidable formalism for complex objects

2 Wide range of structure-based classes3 DLV-based implementation exhibits a significant speedup4 Evaluation over ChEBI ontology revealed modelling errors


Future directions

SMILES-based surface syntaxDetect subsumptions between classesExtensions with numerical datatypesClassification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF

Thank you! Questions?!?

8


Results1 Expressive and decidable formalism for complex objects2 Wide range of structure-based classes

3 DLV-based implementation exhibits a significant speedup4 Evaluation over ChEBI ontology revealed modelling errors


Future directions



8


Results1 Expressive and decidable formalism for complex objects2 Wide range of structure-based classes3 DLV-based implementation exhibits a significant speedup

4 Evaluation over ChEBI ontology revealed modelling errors


Future directions



8


Results1 Expressive and decidable formalism for complex objects2 Wide range of structure-based classes3 DLV-based implementation exhibits a significant speedup4 Evaluation over ChEBI ontology revealed modelling errors


Future directions



8




Future directionsSMILES-based surface syntax

Detect subsumptions between classesExtensions with numerical datatypesClassification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF


8





∧5i=1hasAtom(x, yi) ∧ c(y1) ∧ o(y2) ∧ o(y3) ∧ c(y4)∧

double(y1, y2) ∧ single(y1, y3) ∧ single(y3, y4) ∧ single(y1, y5)→ carboxylicEster(x)



8





define carboxylicEstersome hasAtom SMILES(COC(= O)[∗])end.



8




Future directionsSMILES-based surface syntaxDetect subsumptions between classes

Extensions with numerical datatypesClassification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF


8




Future directionsSMILES-based surface syntaxDetect subsumptions between classesE.g., Carboxylic ester is an organic molecular entity

Extensions with numerical datatypesClassification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF


8




Future directionsSMILES-based surface syntaxDetect subsumptions between classesExtensions with numerical datatypes

Classification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF


8




Future directionsSMILES-based surface syntaxDetect subsumptions between classesExtensions with numerical datatypesE.g., Small molecules if they weigh less than 800 daltons

Classification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF


8




Future directionsSMILES-based surface syntaxDetect subsumptions between classesExtensions with numerical datatypesClassification of complex biological objects

Integration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF


8




Future directionsSMILES-based surface syntaxDetect subsumptions between classesExtensions with numerical datatypesClassification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .

Mapping from our logic to RDF


8




Future directionsSMILES-based surface syntaxDetect subsumptions between classesExtensions with numerical datatypesClassification of complex biological objectsIntegration with Protégé, Bioclipse, JChemPaint,. . .Mapping from our logic to RDF


8

Ontology-Based Classification of Molecules: a Logic Programming Approach

Technology

Transcript of Ontology-Based Classification of Molecules: a Logic Programming Approach