caGrid Version 0.5 Reference Implementation RProteomics caBIG Architecture Workspace Face to Face Georgetown University August 16th -18th, 2005

1
caGrid Version 0.5 Reference ImplementationRProte
omicscaBIG Architecture Workspace Face to
FaceGeorgetown UniversityAugust 16th -18th,
2005

• Patrick McConnell
• Duke Comprehensive Cancer Center
• patrick.mcconnell_at_duke.edu
• Shannon Hastings
• Ohio State University
• hastings_at_bmi.osu.edu

2
Outline

• High Level Overview of Proteomics
• Data Model
• Project Architecture
• Process of getting to Silver level compliance
• Functionality Exposed to Grid
• Process of Grid Enablement
• Demo/Screenshots
• Lessons Learned / Technical Difficulties / Wish
  List
• Acknowledgements

3
Proteomics Overview

• Goal
• Find biomarker
• Build predictive model
• Proteins are split into peptide fragments
• Mass is measured by time-of-flight (TOF)
• Mass of peptides can be used to identify
  proteins
• Peptides can undergo a second MS to help
  identification

http//www.appliedbiosystems.com/catalog/myab/Stor
eCatalog/products/CategoryDetails.jsp?hierarchyID
101category3rd112051trailno
4
Proteomics Data

• A modest study can be on the order of 10 GB of
  data

5
Project Overview

• RProteomics is a development project in the
  Proteomics SIG of the ICR Workspace
• Developing analytical routines for proteomics
  data
• Denoising, background removal, peak
  identification, spectral alignment,
  normalization, peptide quantitation
• Focus is on analytics
• NOT databases, LIMS, protein identification
• RProteomics is a critical step in the proteomics
  pipeline
• LIMS -gt repository -gt RProteomics -gt
  classification -gt protein identification
• RProteomics provides integration
• Q5 classification has been integrated

6
Statistics Background Removal
7
Statistics Denoising
8
Statistics Spectral Alignment
9
Statistics Protein Quantitation
10
Data Model

• mzXML
• Encodes raw spectra data (mz-intensity pairs)
• Some metadata about instrumentation
• Utilizes base64 encoding for binary data
• scanFeatures
• Encodes analysis results as a set of features
• Some metadata about the experiment
• Utilizes base64 encoding for binary data
• Service parameters
• JpegImage
• Lsid
• WindowSize
• ThreshholdMultiplier

11
Project Architecture
12
Project Architecture
13
Process of getting to Silver level compliance

• Programming and messaging interfaces
• Apache Axis for web services
• Wrapped functionality with Java interfaces that
  made sense
• Vocabularies, terminologies, and ontologies
• Data elements
• Wrote tool for XML Schema to XMI conversion
• Manually curated UML
• Went through semantic connecting process
• Information models
• XML Schema to begin with, so information models
  were easy

14
Functionality Exposed to the Grid

• Analytical service no security requirements
• Discuss its input and output and what it does
  scientifically
• Functionality to be exposed
• 20 more statistical methods
• Data access methods, translation methods

15
Process of Grid Enablement

• Process
• Creation/extraction of data types using XML
  Schema
• Upload data types into caGrid GME
• Use the Analytical Toolkit Portal to create and
  modify grid service interface.
• Implement the server stub that is generated by
  making the appropriate calls into the original
  non-grid-enabled RProteomics application.
• Compile, and deploy.

16
Demo and/or Screenshots

• Demonstration of RProteomics GUI with grid
  functionality

17
Lessons Learned / Technical Difficulties / Wish
List

• Think grid from the beginning
• Have an idea what the service interface will be
  ahead of time
• Wrap parameters with objects
• Technology is complex
• XML, Schema, CDEs, Globus, Web Services, etc.
• Installation is complex
• Have to have working knowledge of Tomcat, Axis,
  Ant, environment variables, etc.
• Need to have compatible versions of each
  component, esp. Java 1.4.2_04
• Wish list
• Wizard for grid-enabling existing code
• Documentation of every aspect of installation and
  functionality
• Clone Shannon for each development project

18
Lessons Learned / Technical Difficulties / Wish
List

• Starting with a non-grid-enabled application
  which has been tested and is stable made wrapping
  it to a grid service easier to debug.
• Need a standard mechanism for dealing with large
  data objects.
• Some sort of lazy loaded object/pointer would be
  sufficient.
• Integration of toolkit portal into some standard
  IDEs might make development even easier.

19
Acknowledgements

• Duke, ICR Developer
• Patrick McConnell, Project lead
• Richard Haney, Architect and developer of
  statistical systems
• Salvatore Mungal, Middle-tier Java developer
• Mark Peedin, Database developer
• Northwestern University, Collaborator
• Simon Lin, Proteomics domain expert
• Oregon Health Sciences University, ICR Adopter
• Shannon McWeeney
• Veena Rajaraman
• University of Pennsylvania, ICR Adopter
• David Fenstermacher
• Craig Street
• University of North Carolina, Collaborator
• Cristoph Borchers, Proteomics scientist

• OSU, caGRID Team
• Shannon Hastings
• Scott Oster
• Stephen Langella
• Tahsin Kurc
• Joel Saltz
• Architecture
• Arumani Manisundaram
• Avinash Krishnakant
• VCDE
• Brian Davis, Workspace Lead
• George Komatsoulis, VCDE lead
• Claire Wolfe, VCDE curator
• Salvatore Mungal, VCDE mentor