realism for scientific ontologies
DESCRIPTION
Science aims to develop an accurate understanding of reality through a variety of rigorously empirical and formal methods. Ontologies are used to formalize the meaning of terms within a domain of discourse. The Basic Formal Ontology is an ontology of particular importance in the biomedical domains, where it provides the top-level for numerous ontologies, including those admitted as part of the OBO Foundry collection. The Basic Formal Ontology requires that all classes in an ontology are actually instantiated in reality. Despite the fact that it is hard to show whether entities of some kind exist or do not exist in reality (especially for unobservable entities like elementary particles), this criterion fails to satisfy the need of scientists to communicate their findings and theories unambiguously. We discuss the problems that arise due to the Basic Formal Ontology’s realism criterion and suggest viable alternatives.TRANSCRIPT
Realism for Scientific Ontologies
Michel Dumontier and Robert Hoehndorf
Carleton University and European Bioinformatics Institute
6th International Conference on Formal Ontology in InformationSystems
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 1 / 34
Introduction
Introduction
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 2 / 34
Introduction
IntroductionSummary
Terms in an ontology should correspond to instances in reality.[...] Ontologies consist of representations of types in reality.
realism criteria for biomedical ontologies: all classes must be based on(Aristotelian) universals
no adequate representation of theories and hypotheses
no adequate representation of plans and descriptions
not suited for scientific discourse
harmful for many applications of ontologies in science
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 3 / 34
Introduction
IntroductionWhat is science?
Science is the concerted human effort to understand, or tounderstand better, the history of the natural world and how thenatural world works, with observable physical evidence as thebasis of that understanding (E.O. Wilson).
Science is not truth.
Science is not certainty.
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 4 / 34
Introduction
IntroductionThe scientific method
collection of data through observation
formulation of hypotheses and theories
inference and prediction
experimentation
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 5 / 34
Introduction
IntroductionThe role of ontology
In each step of the scientific method we require ontology:
to describe and classify collected data,
to formulate and express hypotheses and theories
to infer and predict outcomes,
to collect evidence through experimentation, e.g., to find predictedinstances of a type.
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 6 / 34
Ontologies in biomedicine
Biomedical ontologiesDatabase annotation
End of 1990s: genome projects
Large number of homologous genes
Data available and annotated in model organism databases
Different terms used to describe gene functions
Problem: integration and communication problem
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 7 / 34
Ontologies in biomedicine
Biomedical ontologiesGene Ontology
three branches: cellular components, biological processes, molecularfunctions
vocabulary organized as directed acyclic graph
edges labelled is-a, part-of, regulates
textual definitions for all terms
goal: integration of model organism databases
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 8 / 34
Ontologies in biomedicine
Biomedical ontologiesOpen Biomedical Ontologies (OBO)
about 100 domain ontologies
designed as vocabularies for database interoperability
common principles: openness, common syntax, definitions anddocumentation, orthogonality
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 9 / 34
Ontologies in biomedicine
Biomedical ontologiesWhich problem do we want to solve?
ontologies (in biology) intend to solve a communication problem
database integration by defining the terms used in databases
support scientific discourse
specify the meaning of terms in a vocabulary
terms refer to observations, hypotheses, theories, predictions andexperiments
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 10 / 34
Ontologies in biomedicine
Biomedical ontologiesOBO Foundry
“high-quality” ontologies for science
Relationship Ontology and Basic Formal Ontology (BFO)
“Terms in an ontology should correspond to instances in reality.”
“Ontologies consist of representations of types in reality.”
collection of realist ontologies
What is a realist ontology?
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 11 / 34
Realism
RealismTypes of realism
Realism does not say how things are but only that there is a waythat they are. [Searle]
Realism about the world
Modal realism
Scientific realism
Realism about universals
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 12 / 34
Realism
RealismTypes of realism
Realism does not say how things are but only that there is a waythat they are. [Searle]
Realism about the world
Modal realism
Scientific realism
Realism about universals
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 13 / 34
Realism
RealismRealism about universals: Aristotle
universals exist independently from minds
universals exist in things (in re)
something identical in all instances
dependent on particulars
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 14 / 34
Realism
RealismBFO and OBO Foundry
Realism in BFO and OBO Foundry: all classes in an ontology must bebased on Aristotelian universals.
no inconsistent classes
no classes which have no instances in “reality”
only represent types in “reality”
“ontology as reality representation” (Smith, 2004)
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 15 / 34
Realism
RealismScientific theories: the Standard Model
Standard Model in particle physics:
theory of the strong, electromagnetic and weak interaction betweenelementary particles
consistent with most high-energy physics experiments
predictive power: W and Z bosons, gluon, top and charm quarks,decay of Z bosons
Hypothetical and unconfirmed: Higgs boson, graviton, magneticphoton, X boson, Y boson, ...
the meaning of Higgs boson (X, Y boson, graviton, ...) is undisputed
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 16 / 34
Realism
RealismScientific theories: Higgs boson
Electron can be a class in OBO Foundry/BFO
Higgs boson cannot be a class in an OBO Foundry/BFO basedontology until instances are found
theories and hypotheses about Higgs boson exist in “reality”
How can the ontology of the Standard Model be made available forscientists?
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 17 / 34
Realism
RealismInformation, propositions, theories
Information Artifact Ontology (IAO) aims to represent:
documents, texts, data files
parts of documents: conclusions, introductions
numerals
scientific discourse
theories, propositions, statements
aboutness
ICE v ∃is about.Entity
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 18 / 34
Realism
RealismInformation
IAO: ICE v ∃is about.Entity
Entity (BFO) is equivalent to Particular
reify classes to refer to information about types
use BFO’s Generically Dependent Continuant class(BFO’s) universals are generically dependent on their instances
HBT v ∃is about.HiggsBoson
HBT v ∃is about.{HiggsBoson}no instances of HiggsBoson in “reality”
therefore
no HiggsBoson class in BFO/OBO Foundryno HBT class definition in BFO/OBO Foundry
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 19 / 34
Realism
RealismInformation
IAO: ICE v ∃is about.Entity
Entity (BFO) is equivalent to Particular
reify classes to refer to information about types
use BFO’s Generically Dependent Continuant class(BFO’s) universals are generically dependent on their instances
HBT v ∃is about.HiggsBoson
HBT v ∃is about.{HiggsBoson}no instances of HiggsBoson in “reality”
therefore
no HiggsBoson class in BFO/OBO Foundryno HBT class definition in BFO/OBO Foundry
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 19 / 34
Realism
RealismInformation
IAO: ICE v ∃is about.Entity
Entity (BFO) is equivalent to Particular
reify classes to refer to information about types
use BFO’s Generically Dependent Continuant class(BFO’s) universals are generically dependent on their instances
HBT v ∃is about.HiggsBoson
HBT v ∃is about.{HiggsBoson}
no instances of HiggsBoson in “reality”
therefore
no HiggsBoson class in BFO/OBO Foundryno HBT class definition in BFO/OBO Foundry
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 19 / 34
Realism
RealismInformation
IAO: ICE v ∃is about.Entity
Entity (BFO) is equivalent to Particular
reify classes to refer to information about types
use BFO’s Generically Dependent Continuant class(BFO’s) universals are generically dependent on their instances
HBT v ∃is about.HiggsBoson
HBT v ∃is about.{HiggsBoson}no instances of HiggsBoson in “reality”
therefore
no HiggsBoson class in BFO/OBO Foundryno HBT class definition in BFO/OBO Foundry
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 19 / 34
Realism
RealismInformation
HiggsBoson may not be an Aristotelian universal.
Electron probably is.
Consequence
only experimentally supported predictions of the SM can be representedin BFO/OBO Foundryonly information about supported predictions
Can experiments be described using BFO/OBO Foundry?
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 20 / 34
Realism
RealismExperimentation and interpretation
Ontology of Biomedical Investigations (OBI) aims to represent:
investigation, experimentation, documentation
scientific assays
plans and descriptions
outcomes and measurements
interpretation of results
But: to describe experimentation and interpretation, we must be able todescribe what the experiment and interpretation is about.
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 21 / 34
Realism
RealismExperimentation and interpretation
Ontology of Biomedical Investigations (OBI) aims to represent:
investigation, experimentation, documentation
scientific assays
plans and descriptions
outcomes and measurements
interpretation of results
But: to describe experimentation and interpretation, we must be able todescribe what the experiment and interpretation is about.
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 22 / 34
Realism
RealismPlans
[I]t would be an error to include in a scientific ontology of drugsterms referring to pharmaceutical products which do not yet(and may never) exist, solely on the basis of plans anddescriptions. Rather, such terms should be included precisely atthe point where the corresponding instances do indeed exist inreality [...] [Smith et al., 2006]
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 23 / 34
Realism
RealismLimiting research results
De Silva et al. What is the smallest saturated acyclic alkane that cannotbe made? J. Chem. Inf. Model, 45(1):81–87, 2005.
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 24 / 34
Realism
What are chemical structures and their relations?Original example: Hastings et al., 2010
the CU, which has instances individual chemical entities
BetaFarnesene v FarneseneBetaFarnesene ≡ ∃hasSpecification.BetaFarneseneCG
the connectivity, which exists in many individual molecules
BetaFarneseneConnectivityME v ME
the chemical graph representation of the connectivity, which is aninformation artifact, concretized in a particular file on the computer
BetaFarneseneCG v InformationContentEntityBetaFarneseneCG v ∀isAbout.BetaFarneseneConnectivityME
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 25 / 34
Realism
What are chemical structures and their relations?Critique
BetaFarnesene ≡ ∃hasSpecification.BetaFarneseneCG
something must have a chemical graph specification to be a betafarnesene
no MEs in BFO
mathematical graphs become irrelevant
no relation between mathematical graphs and molecules
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 26 / 34
Realism
What are chemical structures and their relations?Application to C17
C 17 v Molecule not allowed
C 17CG v ICE
C 17CG v ∀isAbout.C 17ME
∃hasSpecification.C 17CG
C 17 ≡ ∃hasSpecification.C 17CG not allowed
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 27 / 34
Realism
What are chemical structures and their relations?Application to C17
C 17 v Molecule not allowed
C 17CG v ICE
C 17CG v ∀isAbout.C 17ME
∃hasSpecification.C 17CG
C 17 ≡ ∃hasSpecification.C 17CG not allowed
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 27 / 34
Realism
What are chemical structures and their relations?Application to C17
C 17 v Molecule not allowed
C 17CG v ICE
C 17CG v ∀isAbout.C 17ME
∃hasSpecification.C 17CG
C 17 ≡ ∃hasSpecification.C 17CG not allowed
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 27 / 34
Realism
What are chemical structures and their relations?Application to C17
C 17 v Molecule not allowed
C 17CG v ICE
C 17CG v ∀isAbout.C 17ME
∃hasSpecification.C 17CG
C 17 ≡ ∃hasSpecification.C 17CG not allowed
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 27 / 34
Realism
RealismAlternative scientific theories
Perpetual Motion Machine (first kind)
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 28 / 34
Realism
Formal OntologyWhich problem do ontologies solve?
ontologies to support scientific discourse and data integration
contain all terms used within scientific discourse
Unicorn is not used in scientific discourse
Higgs boson, Perpetual motion, impossible molecules, etc. are used inscientific discourse
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 29 / 34
Realism
Formal OntologyAn old solution
include all relevant terms within a domain/application
specify the meaning of terms: how a term refers
(need for possible worlds with contingent natural laws)
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 30 / 34
Conclusions
ConclusionsSummary
BFO/OBO Foundry: all classes in an ontology must have instances in“reality”
applied in some biomedical ontologies
cannot represent unconfirmed hypotheses, predictions, theories
unsuitable as foundation for scientific discourse and investigation
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 31 / 34
Conclusions
ConclusionsSummary
philosophy is useful to support scientific investigations
“You cannot do what you need because Aristotle said so” is not
realism criterion should not be applied
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 32 / 34
Conclusions
Acknowledgements
Heinrich Herre
Phillip Lord
Robert Stevens
Janet Kelso
Frank Loebe
Nico Adams
DietrichRebholz-Schuhmann
Maria Laikata
Janna Hastings
Philippe Rocca-Serra
Susanna-Assunta Sansone
Helen Parkinson
James Malone
Jennifer Fostel
Anika Oellrich
Darius Sulskus
Christoph Grabmueller
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 33 / 34
Conclusions
Thank you!
Robert Hoehndorf (EBI) Realism for Scientific Ontologies FOIS2010 34 / 34