Knowledge Elicitation: Going LowTech

1
Knowledge Elicitation Going Low-Tech
Susanna-Assunta Sansone NET Project
Coordinator EMBL EBI The European
Bioinformatics Institute Cambridge,UK
Ontogenesis Network Manchester, October
30-31st, 2006
2
Knowledge Elicitation Going Low-Tech
Ontogenesis Network Manchester, October
30-31st, 2006
3
Outline

• Rationale behind this work
• The Ontology for Biomedical Investigation (OBI)
• - An example of collaborative ontology building
  effort
• Knowledge elicitation exercises
• An ontology for the nutrition domain
• - Our method and experience

4
EBI and the NET Project

• Contribute to standards initiatives, defining
  minimal descriptors, XML-based exchange
  formats and CVs/ontology
• Technology-driven efforts, e.g.
• Microarray Gene Expression Data Society (MGED)
• Proteomics Standardization Initiative (PSI),
  HUPO
• Metabolomics Standards Initiative (MSI),
  Metabolomics Society
• Genomic Standards Consortium (GSC)
• Collaborative projects (including initiatives
  above plus many others)
• MIcheck, gathering and integrating minimal
  description checklists
• Functional Genomics (FuGE) Object Model and XML
  serialization of the model
• Ontology for Biomedical Investigation (OBI,
  previously FuGO)
• Develop standards-compliant systems to store and
  exchange data
• ArrayExpress (MGED), Pride (PSI), metabolomics
  planned (MSI)
• Collaborate with domain-specific communities
• Nutrition, Environment and Toxicology domains
  (NET Project)

5
OBI Overview
6
OBI Communities

• Current technology-driven communities
• Omics technologies Standards initiatives
• HUPO - Proteomics Standards Initiative (PSI)
• Microarray Gene Expression Data (MGED) Society
• Metabolomics Society Metabolomics Standards
  Initiative (MSI)
• Other technologies Groups around databases and
  networks
• Flow cytometry
• Polymorphism
• In situ hybridization and immunohistochemistry
• Current biology-driven communities
• NERC Environmental Bioinformatics Center (NEBC)
• Generation Challenge Programme (GCP)
• Biomedical Informatics Research Network (BIRN)
• Immunology Database and Analysis Portal
• National Center for Toxicogenomics, NIEHS
• Nutrigenomics Organization (NuGO)

7
OBI Organization

• Coordination Committee
• Representatives of technological and biological
  communities
• - Monthly conferences calls (papers,
  presentations, new groups etc)
• Developers Working Group
• Representatives and members of these communities
• - Weekly conferences calls (hands-on the
  ontology, tackle issues etc)
• Advisory Board
• Advise on high level design and best practices
• Provide links to other key efforts
• - Barry Smith, IFOMIS (NcBIO and OBO Foundry)
• - Frank Hartel, NIH-NCI
• - Mark Musen, Stanford (Protégé Team and NcBIO)
• - Robert Stevens, Manchester Un
• - Steve Oliver, Manchester Un
• - Suzi Lewis, Berkeley Un (GO and NcBIO)
• Documentation and dissemination
• http//fugo.sf.net

8
Nutrigenomics Community - NuGO

• Nutrigenomics is the study of the response of a
  genome to nutrients
• Using transcriptomics, proteomics and
  metabolomics technologies
• In combination with biometrics and clinical
  tests
• Epidemiological studies, intervention studies,
  gut microflora etc
• The European Nutrigenomics Organization (NuGO)
  includes 22 partners and organisations from 10
  European countries
• Aiming to develop and integrate all facets of
  resources
• An ontology would be one of these resources
• Providing semantics for those descriptors
  relevant to the interpretation and of
  nutrigenomics experiments and analysis of the
  data
• Upper-level framework provides semantics for
  those descriptors common to other domains
• gt Collaboration with OBI
• - Lower-level framework provides semantics for
  those descriptors specific to the nutritional
  domain

9
Our Methodology
10
Scope and Scenario

• Identify the scope
• What is the ontology going to be used for?
• What do you want the ontology to be aware of?
• What is the scope of the knowledge you want in
  the ontology?
• Define competency questions
• Are those questions for which we want the
  ontology to be able to provide support for
  reasoning and inferring processes
• Illustrate possible scenarios
• Which investigations have samples treated with a
  high-fat diet?
• Which investigations employ microarray in
  combination with metabolomics technologies?
• List those investigations in which the fasting
  phase has as duration one day

11
Knowledge Elicitation Phase

• The art of questioning
• Phase where the knowledge engineer gathers in
  the form of concepts and relationships between
  concepts what the domain experts understand to
  exist in that domain
• Challenges
• Expertise is socially distributed
• - Experts have different but complementary
  understanding of the domain, playing different
  roles in the organizations
• gt Project managers have the big picture of the
  experiment
• gt Experimentalists have hands-on experience with
  one or many technologies
• Expertise is geographically distributed
• Domain experts are located in 10 European
  countries
• Domain experts express their knowledge in
  natural language.

12
From Natural Language to Concepts
Seven week old C57BL/6N mice were treated with
low-fat diet. Liver was dissected out, RNA
preparedetc.
13
Knowledge Elicitation Methods

• One to one or one to many interaction
• Narrative approach
• Interviews
• Discussion and focus groups
• Survey forms
• Emails
• Diagrams
• Conceptual Maps (CMs)

14
Conceptual Maps (CMs) Informal Artefacts
15
CMs Limitations and Use

• Not computationally enabled
• Greater utility than other forms of knowledge
  representation such as spreadsheets or word
  processor tables
• Quite intuitive and easy to use
• Have a (very) simple semantics
• Domain experts are able to
• - Represent concepts
• - Add definitions and examples
• - Declare relations
• Useful to test the representations in
  decentralised settings
• CMAP-tools version 3.8 as a CM editor
• Freely available
• http//cmap.ihmc.us

16
CMs Knowledge Elicitation Sessions
17
CMs Knowledge Elicitation Sessions

• Initial session(s)
• Move from the narration to a list of concepts
• Define some starting concepts
• Agree on what went where
• Define an early structure of the relationships
  that glue the information together
• Following sessions
• Focus on structural aspects of specific concepts
• Guide the discussion by asking
• How does A relate to B?
• Why do we need A here instead of B?
• How does A impact on B?
• Discuss cardinality issues

18
CMs Knowledge Elicitation Sessions

• Iterative process
• Moving from instances to classes
• Create different levels of abstractions
• Identify is_a and defining the whole/part-of
  relationship

E.g. Define the content of a meal
19
CMs Knowledge Elicitation Sessions

• Caveats
• Accuracy in the definition of terms
• - Create consistent label for the terms
• - Add the context where the term is used
• - Provide examples
• Coherence of the story produced
• Ensure consistency within the narration, as CMs
  are being enriched as iterative process
• Extensibility of the representation
• Add more details to the existing CMs
• Generate new CMs
• Group concepts into higher-level abstractions
• gt Validate these with domain experts
• gt Analyse the models from different angles or
  perspectives

20
The experts need to perceive significant
benefit before they can provide information to
others Dave Randall
21
Explain the Benefits
22
Engage with the Experts Build the Trust
23
Resources and Acknowledgements

• Nutrigenomics domain experts
• Ruan Elliot (Institute of Food and Research,
  IFR)
• Anne-Marie Minihane (Food Science, Reading
  University)
• Their postdocs and technicians
• Other (OBI) experts
• Jennifer Fostel (NIH-NIEHS)
• Norman Morrison (NERC Bioinformatics Center,
  NEBC)
• Funds
• NuGO (EU NoE 503630), and Semantic Mining (EU
  NoE 507505)