Information Intensive Breakout

1
Information Intensive Breakout 3

• Jim Myers, Debbie Gracio, Ron Taylor, Larry Rahn,
  Ray Bair, Tom Allison, Karlo Berket, Lawrence
  Buja,

2
Summary

• Social barriers to collaboratory efforts.
• How can we help Virtual Organizations - in set
  up, and in maintaining their information store
  after a VO goes away.
• Loosely-coupled federation of databases seen
  emerging, tied together through exposed web
  services and ontology use, for information
  sharing.
• Importance of template and workflow storage.
• Examples taken from biology and combustion
  communities.

3

• Journal literature is a hyper-compression of
  data. We need to agree to move to a less
  compressed form for annotation by meta-data and
  explicit tagging of data items.
• How does a community get to a level where
  knowledge, rather than data, is hosted?
• Program managers strategically need to encourage
  standards where they make sense for data formats
  and data conventions. Pure chaos when everyone
  has their own data formats, and interoperability
  becomes the bottleneck.
• Biology is a different funding model (vs, say,
  climate modeling). Only the individual lab is
  extracting the value - this is the model today,
  but it is evolving.

4

• Some experimental raw data is valuable and can
  not necessarily be found again by experiment
  repetition.
• Model data can be re-run over and over again
  (perhaps improving in the process).
• Community needs to be involved in the electronic
  standards to support our programmatic efforts.
  Today we search against publications rather than
  against data sets publicly available within
  databases - lack of standards in experimental
  descriptions.
• Combustion community is stalled because they
  dont know how to share data cultural issue
  around sharing data, and format issues. Changes
  will need to occur in the community to support
  better progress to meet the mission needs of DOE.
  
• Basic Energy Sciences needs to become a stronger
  supporter of this we need to make the
  value-based argument.

5

• Need to build trust within science community so
  that the CS community can respond to the needs
  and changes of the science community based
  around formats and next generation tools to work
  with the data.
• In bio community very difficult to capture the
  context of the experiment success in
  microarray, working towards standards in
  proteomics.
• Is there a general base of information that can
  be pulled out to facilitate capture across
  multiple data sources? Build standards for that
  base, then evolve the standards over time.
• Experience with the data is very important before
  a standard can be defined. Workshops within
  community organizations in the biology area are
  starting to work towards standards.
• Portal provides mechanism to share information
  whats the future of this infrastructure to allow
  us to continue to build communities?

6

• Portal usage is based on community needs Group
  spaces vs. federated databases, e.g., molecular
  biology.
• Strawman portal interface is stood up by a
  virtual organization with unique identifiers to
  databases. The virtual org wont own the data,
  which may be archived in a variety of locations
  that expose multiple web-based services that can
  work on the data.
• Central web service registry (BioMoby based?).
  For biology, perhaps at NCBI.
• Hypothesis stand-up of a VO is a key
  capability, with branding, control over policies,
  but with little investment in services, hardware,
  etc. The VO would create an instance of a portal
  to have a presence within their own community,
  for little investment.
• Concept of mingling data across databases
  analogy of a Sourceforge for data, rather than
  program development. Groups can use common web
  services through local registries.
• Data sources must have a value independent of the
  technology lasting value in the original data.
  There is a need to maintain the provenance of the
  metadata, data, and model results.

7

• Strawman concept VO written to support multiple
  services, using similar look and feel. Component
  based, with a dynamic model for how it operates,
  based on policy. The component communicates its
  own semantics so that the developers/users dont
  worry about implementations.
• Can assume that there is a translator between
  different data sources, between different
  ontologies (may be noop).
• Assembly and propagation of ontologies are
  required to make electronic data items useable as
  knowledge. Need ontology translation process to
  tie ontologies together.
• Biology community has multiple groups building
  ontologies. A community based ontology is the
  end product that all agree to work towards.
  Individual labs need to participate in ontology
  creation, and tie data items in local database to
  appropriate nodes in appropriate ontologies.

8

• Want to be able to define an ontology that works
  for me, and then extend by adding other
  ontologies that add meaning to my data.
• Want a suite of software tools that codes
  assumptions, hypothesis, and results into data
  and then connects with collaborators to share
  knowledge in a useable form.
• Journals are moving towards standardized data
  formats within the bio community. Journals are
  requiring deposit into public databases eg. NCBI,
  EBI, NIST. How will people access the
  experimental data in such archives?
• NIST model requires review/curation of data prior
  to archival which then also supports review for
  the journal. Data is then hosted in XML form
  within the NIST database. implemented for one
  specific community today, NIST wants to extend to
  more communities.
• Projected bio future huge number of databases
  available on web will be data mined/knowledge
  mined via web services that are available to
  access each database, the services placed in a
  centralized registry. Published ontologies will
  be used within databases, interpretation of terms
  based on ontologies, loose federation of
  databases adopting unique identifiers (eg. Life
  Sciences identifier). Data items exchanged
  across web, tagged with metadata.

9

• The importance of a globally unique identifier
  for data items.
• Microarray public repositories will allow data
  mining to correlate your research data with that
  of other labs. Data mining will be extremely
  valuable, but each database must be designed to
  support such, and correlated ontologies are
  important to support mining across multiple
  databases.
• Allowing the creation and storage of templates is
  important for workflow, as example.
• Capturing workflow would also use of algorithm to
  provide quality control and validation.
• A groups definition of the queries to be posed
  and the ontologies required to support knowledge
  discovery across communities, the need to share
  of workflow definitions and templates of
  procedures this then is a starting point for
  the development of data policies.