Outline

1
(No Transcript)
2
Outline

• Why animal behaviour?
• Why animal behavior?
• Lesson One
• To help structure descriptors of behaviour, an
  ontology is a good thing
• Lesson Two
• When seeking community consensus, a complex
  ontology may be not such a good idea

3
Betty, the tool-making crow
A.A.S. Weir, J. Chappell and A. Kacelnick,
Shaping of hooks in New Caledonian crows, Science
297 (2002) 981.
4
VANQUIS - a computer-aided video analysis tool
5
VANNOTEA an ontology-driven video analysis tool
http//www.metadata.net/filmed
6
SABO development in Oxford, 2003

• In 2003, we developed a draft Standard Animal
  Behaviour Ontology, SABO
• The purpose was to facilitate semantic content
  analysis of animal behaviour videos in a
  principled manner, particularly those made by our
  Oxford colleagues, permitting interoperability of
  the metadata descriptors thus created
• Conceptually, SABO was complex, based on Niko
  Tinbergens four questions
• How did a behaviour develop
• What is its function
• How is it controlled
• What is its evolutionary origins
• SABO was centrally organised around hypotheses
• It sought to make a clear distinction between
• observed behavioural activities (objective
  reality) and
• subjective (albeit expert) human interpretations
  of the purpose or significance of those
  activities

7
SABO development in Oxford, 2003

• Process
• All done without dedicated staff or funding!
• Two students, Chris Wilson and Ruth Dalton, were
  independently set the task of thinking what an
  ontology to annotate animal behaviour videos
  should comprise
• Based on their work, discussions with me, and his
  own additional input, Chris Catton, who had
  extensive experience in making wildlife films for
  the BBC, then wrote the first draft of SABO using
  Protégé
• This was later converted into OWL when the
  Protégé OWL plug-in became available

8
Modelling fact and hypothesis in SABO
9
A fragment from the original SABO ontology
10
The first Cornell Animal Behavior Metadata
Workshop

• Timing A weekend in April 2004, at the
  Ornithology Labs, Cornell University
• Host Jack Bradbury, head of Macaulay Library of
  Natural Sounds
• Participants Around 30 leading international
  academic ethologists, plus Chris Catton and
  myself
• Purpose to promote metadata consistency in
  animal behaviour recordings, and thus enable
  interoperability between data in distributed
  institutions (academic research units, zoos,
  etc.)
• Inputs
• Macaulay Librarys metadata model
• Oxfords draft SABO
• Naïve ideas from Chris Catton and myself about
  the benefits of the Semantic Web and of
  ontologies in place of thesauri
• Enthusiasm and domain expertise on the part of
  the participants

11
My ideas for success criteria in ontology
development

• Based mainly upon my knowledge of the Gene
  Ontology
• Involve international collaboration and consensus
  building
• Use appropriate technologies
• Adhere to acknowledged or emerging international
  standards
• Publicize by means of descriptive papers in
  prominent journals
• Ease of uptake by ensuring simplicity of
  implementation
• Rapid creation of a wide user base to achieve
  critical mass

12
My view of factors mitigating against success

• Development
• The cost of building international collaboration
  - in time and money
• Getting hung up on the details, rather than
  concentrating on getting the basis structure
  correct
• Usage
• Lack of willingness of the community to invest
  the intellectual effort to use the new metadata
  standard consistently
• Vested interests and peoples unwillingness to
  change established working practices
• The Not invented here syndrome

13
Range of content of ABO v1

• Output The Animal Behavior Ontology v1, written
  in Protégé, released June 2004
• Each specific behavioral event had a wide range
  of sub-fields including
• Behavior Type, common name
• Subjects Actor, recipient, observer, subject
  relation, external target
• Structures body parts, external objects
• Communication Signal content, kinematic content
• Spatial relative location, spatial pattern and
  spatial dynamics of each interactant
• Spatial pattern of interactants
• Spatial dynamics
• Contexts experimental, observational
• Other Learning and development implications
  theoretical behavior concepts
• Ambitious is scope, to say the least !

14
A view of the ABO v1 class heirarchy

• Note problems
• lack of consistent syntax
• ambiguities within class names

15
The second Cornell Animal Behavior Metadata
Workshop

• Timing A weekend in September 2005, at the
  Ornithology Labs, Cornell University
• Host Jack Bradbury, head of Macaulay Library of
  Natural Sounds
• Participants A selected subset of those leading
  international academic ethologists who had
  attended the first workshop, plus Chris Catton
  and myself
• Purpose to improve ABO v1 to a point where it
  might be useful
• Inputs
• ABO v1
• A paper from Jack about the need for progress
• Catton and Shottons paper Ontologies for
  Sharing, Ontologies for Use
• My diagram of the community ontology development
  cycle
• My analysis of problems with ABO v1
• My suggestions of how to improve ABO
• Commitment from participants to work hard
  together towards the goal

16
Ontologies for Sharing, Ontologies for Use

• Based on principles that Alan Rector enunciated
  here during a Protégé-OWL workshop a couple of
  years ago, Chris Catton and I wrote a short paper
  with the above title that we shared with the good
  folk at Cornell. In essence
• A public, shared ontology should
• be kept small or very small, ideally as a set of
  simple normalised subsumption (is_a) hierarchies
• require minimal ontological commitment
• not import any other ontology i.e. it stands
  alone
• define a limited domain
• Small ontologies are advantageous because they
  are
• easier to manage
• easier to validate with domain experts
• easier to achieve consensus - fewer classes and
  properties to disagree about
• easier to import
• An application level ontology
• imports the public ontology (and public
  ontologies describing other domains)
• restricts the public ontology and thereby enables
  more powerful reasoning

17
The community ontology development life cycle
18
Problems with ABO v1

• I identified the following problems with ABO v1
• It contained a rich mixture of dissimilar terms
  that were not clearly differentiated
• It contained duplications under different
  superclasses, e.g. Play (9.3.8) and Play behavior
  (11.1.4.14).
• It contained logical inconsistencies, e.g. Good
  genes (10.2.5.2.2) is not a type of Sexual
  selection (10.2.5)
• It used vague or ambiguous class names, e.g.
  Extracted measure (2), External object (3.5.3),
  Relevant concepts (10)
• It contained classes such as Color (8.1.4.8),
  Genetic mechanisms (10.9.4.4.1) or Common name
  (11.2) should be part of separate third-party
  ontologies
• Most importantly, it lacked clear distinctions
  between objective descriptions of behavioural
  events (e.g. running, sitting still) and their
  functional interpretations (e.g. prey capture,
  awaiting prey)

19
Proposed improvements

• I suggested dividing the ontology into four
  independent pure is_a hierarchies
• Behavioural Events, describing the ground facts
  about what the animal does running, biting,
  scratching, vocalizing, copulating, shoaling,
  etc.
• Great care must be taken to exclude functional
  interpretations and anthropomorphisms from this
  ontology.
• Behavioural Functions, describing the functions
  attributed by human domain experts to observed
  items of behaviour e.g. acquisition of a mate,
  assertion of dominance, guarding of territory,
  learning, prey capture
• Behavioural Causes, describing the proximate
  causes attributed by human domain experts to
  observed items of behaviour e.g.
• behaviour in response to sensory stimulus (e.g.
  sight of predator)
• behaviour in response to physiological state
  (e.g. hunger)
• Behavioural Models and Theories attributed by
  human domain experts to observed items of
  behaviour for the purpose of bringing theoretical
  coherence to the observed events
• Optimal foraging, inclusive fitness, time
  budgeting, aposomatic signalling, etc

20
What actually happened

• First day Spent analysing shortcomings and
  discussing whether or not to adopt my suggestions
  to simplify ABO into these simpler subsumption
  hierarchies
• Second day Divided into two groups of 8, one to
  look at Behavioral Acts and the other to look at
  Behavioral Functions
• Activities
• Group one pruned out classes from existing
  ontology, leaving just Behavioral Acts
• Group two developed a Behavioral Functions
  hierarchy from scratch
• Two groups then came together and reviewed each
  others work surprising unanimity about
  results
• All unused terms put into an orphan classes group
  for later attention
• Outputs Animal Behavior Ontology (ABO) v2,
  published March 2006 and available at
  http//ethodata.org,
• Behavioural actions
• Behavioural events
• Orphan classes

21
A view of ABO v2

• Note remaining problems
• lack of definitions
• ambiguities within class names
• lack of consistent syntax

22
In the meantime . . .

• The MRC Mammalian Genetics Laboratory at Harwell,
  Oxfordshire, have a programme of individually
  knocking out each gene of the mouse and looking
  at the resulting phenotype
• These knockouts sometimes result in behavioural
  phenotypes
• Georgios Gkoutos, a member of the Ontogenesis
  Network then at that lab, had developed a
  behavioural assay ontology using OBO-Edit
• John Hancock, head of bioinformatics at MRC
  Harwell, expressed interest in working with us to
  combine that ontology with the Animal Behavior
  Ontology
• Joanna Bagniewska, an M. Sc. in Biology student
  who had extensive previous experience of animal
  behaviour research, expressed an interest in
  taking this animal behaviour project forward, and
  has been working on this since January
• My long-term goal is to take the output from her
  work, and with external funding to develop it
  into a more fully specified ontology, by means of
  text mining the corpus of the journal Animal
  Behaviour that the Elsevier has kindly offered to
  make available to us as XML

23
over to Joanna . . .
24
Initial assumptions

• Limited to is_a relation
• Rather small number of nests (max. 4)
• Bare bones of the ontology time constraint!
• Data published on a wiki
• http//ontogenesis.ontonet.org/moin/AnimalBehaviou
  rOntologyDevelopment

25
Starting point

• Cornell ontology version ABOcore0.1.owl
• Number of classes 292
• Very detailed in some parts, very broad in others
• Needed a formatting clean-up

http//ontogenesis.ontonet.org/moin/AnimalBehaviou
rOntologyDevelopment
26
Animal Behavior Ontology

• Version ABOv0.1.owl
• Downloadable from the wiki
• Edited following the guidelines of Barry Smith
• Class names are singular nouns
• Class names are not capitalized
• Class names are limited to single terms
  (eliminating /, or etc.)

http//ontogenesis.ontonet.org/moin/AnimalBehaviou
rOntologyDevelopment
27
Animal Behavior Ontology

• Current version 349 classes
• American spelling preserved (more universal)
• Two separate ontologies behavioral acts and
  functions.
• Data from literature, communication, personal
  experience.

http//ontogenesis.ontonet.org/moin/AnimalBehaviou
rOntologyDevelopment
28
Encountered difficulties

• territoriality is_a behavioral_function BUT
• patrolling_behavior is_a territoriality_aspect

29
Encountered difficulties

• is_a relation very limiting
• Act vs. function
• act grouping
• function huddling
• OR
• act huddling
• function temperature regulation
• volitional acts?
• Differs with species, context, etc.

30
Animal Welfare Ontology

• Startup ontology!
• Referring to animals under human husbandry
• Number of classes 265
• Three separate ontologies.

http//ontogenesis.ontonet.org/moin/AnimalBehaviou
rOntologyDevelopment
31
Eventual goal

• MRC Harwell ontology of mice welfare
• More relations, more precision
• Adding factors like assays, units etc.

32
Encountered difficulties

• Limited number of relations not precise enough
  for some descriptions
• Biased wording
• Differs with species, context, etc.
• Trade-off between user-friendliness and precision
• Creating is different from using!

http//ontogenesis.ontonet.org/moin/AnimalBehaviou
rOntologyDevelopment
33
Communicating with the public help needed!
34
end