EXPERIENCES IN BUILDING AN ONTOLOGYDRIVEN IMAGE DATABASE FOR BIOLOGISTS

1
EXPERIENCES IN BUILDING AN ONTOLOGY-DRIVEN IMAGE
DATABASE FOR BIOLOGISTS
2
Outline

• Why are images important?
• What is the BioImage database?
• Why use a semantic web architecture?
• Lessons and research questions

3
Why are biological images important in the
post-genomic age?

• Images are semantic instruments for capturing
  aspects of the real world, and form a vital part
  of the scientific record, for which words are no
  substitute
• In the post-genomic world, attention is now
  focused on the organization and integration of
  information within cells, for functional analyses
  of gene products
• In a month a single active cell biology lab may
  generate between 10 and 100 Gbytes of
  multidimensional image data

4
Images are complex

• An image database must be able to store original
  images in any digital format currently available
  or yet to be invented, including multi-channel 3D
  images, multi-channel videos, etc.

5
The need for image databases

• The value of digital image information depends
  upon how easily it can be located, searched for
  relevance, and retrieved
• Detailed descriptive metadata about the images
  are essential
• Without them, digital image repositories become
  little more than meaningless and costly data
  graveyards
• Despite the growth of on-line journals that
  permit the inclusion of media objects, few of
  these resources are freely available, and those
  that are are difficult to locate and are not
  cross-searchable
• There is thus a need for a free publicly
  available image database with rich
  well-structured searchable metadata
• The BioImage Database seeks to fulfil that need

6
This view has a growing acceptance
7
What metadata?

• Image acquisition (who took the original
  micrograph, where, when, under what conditions,
  for what purpose, etc.)
• The media object itself (source and derivation,
  image type, dynamic range, resolution, format,
  codec, etc.),
• The denotation of the referent (e.g. the name,
  age and condition of the subject),
• Connotation of the referent (the images
  interpretation, meaning, purpose or significance,
  its relevance to its creator and others, and its
  semantic relationship to other images).
• Field aspects of the real world that cannot
  conveniently be attached to any particular object
  (e.g. variations of illumination intensity or
  chemo-attractant concentration across the field
  of view of a light microscope image).
• Sequences of change where there is a need to
  preserve the concept of object identity in the
  face of radical spatio-temporal changes in
  appearance.

8
Why use a semantic web architecture?

• Traditional relational databases dont meet our
  needs
• Image data is complex, layered, and difficult to
  model
• Images are searched primarily through their
  metadata
• Metadata is time consuming and difficult to
  obtain
• Ontologies offer the promise of better retrieval
  accuracy through linking to instances in an
  ontology, rather than attempting to process free
  text.
• Ontologies offer the promise of easy
  inter-operability with other systems

9
The BioImage Ontology
10
Lessons learnedPerformance, scalability

• Database retrieval is slower than a traditional
  database would be
• Scalability remains to be tested (true for all
  semantic web software)
• Query languages (RDQL) are immature when compared
  to SQL
• Parsing RDF is hard and slow (RDF-ABBREV output
  of the Jena parser is unreliable and the
  unstriped format requires multiple passes to
  create XML that can easily be transformed to HTML)

11
A problem with ontologies?

• The volume of data generated in the Life Sciences
  is now estimated to be doubling every month
• Already people look less and less at the raw
  scientific data (unless they are their own
  results)
• As this volume of data accumulates, few if any of
  us will have the time or the mental capacity to
  assimilate new data, structure them in a
  meaningful way and extract information, without
  first processing the data through an ontology or
  some other similar machine-based organisational
  aid
• THE ONTOLOGY WILL BE WRONG! (or we should all
  pack up and go home)

12
Paradigm shifts

• Our human understanding of an area of science is
  never static, but is constantly being revised by
  new research
• Such revisions in understanding are either
  evolutionary (incremental), following the
  progressive discovery of more and more detail,
  interpreted according to the prevailing paradigm,
  or revolutionary, when the prevailing paradigm is
  overthrown by another
• How do paradigm revolutions succeed?
• "A new scientific truth does not triumph by
  convincing its opponents and making them see the
  light, but rather because its opponents
  eventually die, and a new generation grows up
  that is familiar with it"
• (Max Planck, 1949)

13
Factors preventing evolution

• Ontology builders are monks (and nuns) - led by
  an abbot, a relatively senior domain expert
  likely to be committed to encapsulating the
  dominant paradigm
• Substantial problems confront any newcomers
  wishing to contribute, since ontology building is
  time-consuming and expensive
• Since an ontology expresses the community
  consensus, there will be massive social pressures
  against change
• If large volumes of data have already been
  encoded using an existing ontology, this will
  make it difficult to introduce change
• The first ontology in a domain may assume a
  monopolistic position that becomes unassailable,
  even if it has universally acknowledged
  weaknesses
• Ontologies are unlikely to evolve in response to
  the same market forces that drive the development
  of applications software

14
Encapsulating the dominant paradigm

• Imagine a section of an ontology describing the
  development of adult mammalian bone marrow and
  brain, constructed according to the pre-1980
  dominant paradigm that bone marrow develops from
  mesoderm, while brain develops from
  ectoderm

15
An example of paradigm evolution

• Subsequently, adult mouse brain was found to
  contain haemopoietic stem cells
• Bartlett (1982) hypothesised that these cells
  developed from foetal haemopoietic cells that
  entered the brain tissue before the barrier was
  established
• This challenge to the dominant paradigm that
  brain tissues are derived exclusively from
  ectoderm can be accommodated by extending the
  graph

16
An example of paradigm revolution

• More recently, Brazelton et al. (2000) claimed
  that haemopoietic stem cells from adult bone
  marrow can develop into neural cells in adult
  mouse brain
• If true, this result overthrows the paradigm that
  neuronal cells can only develop from embryonic
  ectoderm, requiring a new ontology incompatible
  with the old
• This new ontology is no longer an extension of
  the previous one, since neural cells no longer
  develop only from foetal neuroepithelium

17
A way forward using Named Graphs in RDF (and
OWL?)

• In response to considerable frustration and
  confusion within the RDF community about the best
  method of reifying RDF statements, Jeremy Carroll
  et al. proposed an extension to RDF

18
Thanks and acknowledgements

• David Shotton and Simon Sparks for BioImage
  developments (http//www.bioimage.org)
• John Pybus, our computer systems manager, for
  keeping us running in spite of the problems
• Liz Mellings for unbounded patience inputting
  data and testing
• The European Commission for funding the BioImage
  Project (EC IST 5th Framework Contract
  2001-32688 ORIEL Online Research Information
  Environment for the Life Sciences
  http//www.oriel.org)

19
End