NCICB Pathway Interaction Database

1
NCICB Pathway Interaction Database

• 19 February 2004

2
Problem

• Most collections of pathway data are not
  computable
• cant merge pre-defined pathways
• cant find 1st-degree, 2nd-degree downstream
  interactions, ...
• cant propagate hypothesized effect of anomalies
• cant mix data sources (high-throughput curated
  data)
• Prototype Pathway Interaction Database
• http//cmap.nci.nih.gov/PW
• See if sufficient data is available
• Experiment with representations
• See what users want
• Try out algorithms

3
Goals

• Merge data from multiple sources
• Construct cause/effect networks dynamically
• Offer rich set of query options
• molecule, interaction, pathway,
  successor/precedessor, prune, connect, ...
• Associate values (e.g. expression level) with
  molecules propagate across the network
• Offer text and graphic output
• Scale to large network
• currently, graphic output is limiting factor

4
NOT Among Current Goals

• Custom pathway layouts or artistic renderings
• Encyclopedic or unstructured knowledge about
  interactions
• Representing multi-cellular processes
• Controlled vocabulary for pathway names

5
Representation (1 of 2)

• Pathway directed graph
• node molecule or event or condition
• edge role of molecule/condition in an event
• interaction event its connected
  molecules/conditions
• Molecule type
• protein complex compound rna
• or families (e.g. EC_2.7.7.15 includes PCYT1A,
  PCYT1B)
• Event type
• reaction modification transcription
  translocation
• or any GO BP type

6
Representation (2 of 2)

• Condition type
• any GO BP type
• Role type
• input output agent inhibitor
• or any GO MF type
• Molecule location
• any GO CC type
• specified at point of use
• Posttranslational modification
• abstract terms (e.g. active)
• specified at point of use

7
Accessing Data

• http//cmap.nci.nih.gov/PW
• HTML forms
• Command language
• file upload
• post to web service
• Rudimentary data input tool

8
Query Options

• Basic query
• by set of molecules
• by set of predefined pathways
• Query modifiers
• upstream/downstream interactions
• connect molecules (arbitrary shortest connection)
• prune molecules
• collapse distinctions among posttranslational
  mods
• collapse graph by removing process nodes

9
Output Options

• HTML
• individual molecule interaction pages
• Text
• native XML
• SBML
• Graphic
• GIF
• SVG

10
Example ATM and Successors
11
Current Contents (all human)
Interaction Types
12
Technical Details

• Oracle (25 tables)
• no Oracle-specific features used
• Perl (10K lines)
• query parsing
• retrieval from DB
• graph operations value propagation
• output
• a bit of Python HTML
• For layout, dot tool (open source, part of
  GraphViz)
• http//www.research.att.com/sw/tools/graphviz/

13
Related Projects

• pathway interaction data
• BIND
• http//www.blueprint.org/bind
• Genome Knowledge Base
• http//www.genomeknowledge.org
• pathway construction/analysis tools
• Cytoscape
• http//www.cytoscape.org

14
Possible Future Steps

• Create caBIO objects for Pathway Interaction
  Database 4 mos.
• Standardize on XML representation 1 mo.
• BioPAX likely to be more appropriate than SBML
• Implement standard (e.g. SOAP) web services
  interface 1 mo.
• Remove Zope/Python dependencies and package for
  installation 2 mos.
• Further data acquisition open-ended