Chemical Informatics Data Services

1
Chemical Informatics Data Services

• Building cyber-infrastructure to support research
  in small molecule chemistry.

2
On-Demand Data, Not On-Demand Supercomputing

• Cyber-infrastructure (CI) is often discussed in
  terms of hardware, middleware, networking, and so
  forth.
• But its the data
• Can we make data products for the community using
  HPC?
• See for example the Earth Systems Grid
• Model use computational CI to make public
  community data and associated Web services (Data
  Centric CI)

3
Data-Deluged Small Molecular Chemistry

• We are at the beginning of a new era in publicly
  available information on drug-like molecules.
• Small means molecular weight lt 500
• The NIH PubChem database provides data on over
  8,000,000 molecules.
• Free, open, community driven
• Both web browser and web service interfaces.
• The NIH is funding High Through Put Screening
  Centers to analyze gt 300,000 drug like molecules.
  
• Data is deposited in PubChem.
• It is free, open to all researchers.
• NIH PubMed provides access and advanced search on
  1,000s of medical an related journals.
• Now this is cyber-infrastructure

4
Chemical Informatics and Cyber-Infrastructure

• The NIH also sees chemical informatics as key to
  enabling scientific discoveries with PubChem,
  PubMed, and other services.
• Indiana University is funded by NIHs ECCR
  program to research these issues.
• Geoffrey Fox, PI
• Our general program combine expertise in
• Chemical informatics and computational chemistry.
• Grids and high performance computing
• Web Services and workflows.
• Build cyber-infrastructure to enable our own
  research and research of others.

5
Web Services and Workflows

• Web services provide simple developer
  interfaces/message types to complicated remote
  data and computing resources.
• Workflows connect services together.
• Models scientific usage scenarios

Taverna workflow connects remote services.
6
Web Services
7
Chemical Informatics on Big Red
MOAD Database
PubMed Database
OSCAR Text Analysis
Toxicity Filtering
Cluster Grouping
Docking
PubChem Database
Initial 3D Structure Calculation
NIH PubChem Database
NIH PubChem Database
Molecular Mechanics Calculations
Product databases are wrapped with Web service
interfaces and are suitable for inclusion in
Taverna workflows.
Quantum Mechanics Calculations
IUs Varuna Database
POV-Ray Parallel Rendering
8
PubMed and Supercomputing

• PubMed annually indexes 100,000s of medical
  journal abstracts from gt 4,800 journals.
• How can we mine this literature to find
  interesting molecules?
• OSCAR3 is a chemistry-specific text mining tool
  developed by Peter Murray Rusts group at
  Cambridge University.
• Is able identify chemical names in text, can
  extract SMILES (that is, string representations
  of chemical structures.
• SMILES are starting point for many interesting
  workflows.
• See www.chembiogrid.org/wiki for several examples.

9
OSCAR3 Mining on Big Red

• We have used Big Red to mine the entire set of
  PubMed abstracts for chemical structures.
• 555,007 PubMed abstracts of 2005 2006 (part)
  initially run.
• Currently working on the entire catalog.
• OSCAR3
• Extracts chemical information from text and
  produces an XML document highlighting the
  chemical information
• SMILES extraction
• Extracting SMILES elements from OSCARs XML
  output files
• Use this to drive docking and molecular modeling
  applications.

10
Results from Big Red Demo
Final HTML pages
11
Big Red and PubChem

• We are using Big Red to calculate
• 3D structures of 8,000,000 molecules in PubChem
• Molecular Modeling, precursor to quantum
  calculations
• Precise structures and molecular orbitals using
  GAMESS
• Precursor to specific property calculations.
• Preliminary protein docking of all 1,000,000
  drug-like molecules
• Using MOAD protein data base from Heather Carlson
  (UMich) (over 10,000 curated proteins).
• Precursor for more sophisticated docking
• Our goal is to provide data service
  cyber-infrastructure
• These are 3 databases wrapped as 3 web services
  and available through 3 web interfaces.
• These enable our own research
• See Mookie Baiks presentation at the bandwidth
  challenge
• These may be incorporated into workflows
• But they are being made public as well.

12
(No Transcript)
13
Status

• OSCAR3 is available as a Web Service
• 3D Structure Web service is available.
• 85,000 structures now
• 8,000,000 soon
• We omit inorganic molecules, molecules with
  transition metals, etc. (about 1.0-1.5 of
  PubChem)
• These are linked to local copy of PubChem.
• Web service and user interfaces available.
• Protein docking results
• 1,000,000 drug-like molecules in PubChem have
  been docked.
• OpenEyes FILTER finds 983,734
• Heat shock protein (1YC4), Kinases (1R1P, ...)
• Implicated in various cancers, researched by IU
  chemists
• Varuna QM pedigree, structures, and orbitals.
• Postgres database implemented (port of MS access)
  with preliminary interface
• 2,000 QM structure calculations underway.
• User and service interfaces in development
• We roughly estimate 1,000,000 molecules1 Big
  Red Year, so we will need to move to TeraGrid.
• We collect user interfaces to these services in a
  portal.

14
Acknowledgements and More Information

• For more information, see
• www.chembiogrid.org/wiki
• Work done here described by the following people
  
• Kevin Gilbert SMILES to 3D conversions and
  Molecular Modeling calculations
• Rajarshi Guha Protein docking and 3D structure
  web services
• Jake Kim OSCAR3 web services and PubMed mining.
• Mookie Baik QM calculations
• Pulan Yu Varuna web services

15
Additional Slides
16
(No Transcript)
17
CICC Project Information

• Chemical Informatics and Cyberinfrastructure
  Collaboratory is an NIH and MS-funded research
  project to combine the CIs.
• Project web site and more information
• www.chembiogrid.org
• www.chembiogrid.org/wiki
• Team members include
• Computer Science Geoffrey Fox (PI), Dennis
  Gannon, Beth Plale, Marlon Pierce, Yuqing
  (Melanie) Wu, Malika Mahoui, Jake Kim
• Chemical Informatics and Chemistry Gary Wiggins,
  Mu-Hyun (Mookie) Baik, David Wild, Rajarshi Guha,
  Kevin Gilbert
• I have stolen slides and content from these fine
  people.

18
PubDock Docking PubChem

• PubChem contains 8M molecules
• Some are drug like, some are not
• Given a protein we'd like to know what molecules
  will bind to it
• Useful in drug discovery
• Can also be used in non-drug related scenarios
• molecular probes
• imaging

19
PubDock Docking PubChem

• We could
• Dock arbitrary molecules to proteins on the fly
• Dock PubChem and store the results
• Docking PubChem is useful since it allows to
  store and compare results
• We currently select proteins of interest to us
• Heat shock protein (1YC4)
• Kinases (1R1P, ...)

20
Some Numbers

• PubChem has 8M compounds
• We only considered compounds that had between 10
  and 150 heavy atoms
• 7,949,658 compounds
• Many of these compounds are extremely flexible
• Conformer generation takes a long time
• To speed up the process we filtered for drug like
  molecules

21
What Are Drug Like Molecules?

• Used OpenEye's filter program, using default
  settings
• 150 lt MW lt 440
• 10 lt Heavy atom Count lt 25
• -5.0 lt XlogP lt 4.0
• Maximum of 3 rings
• ....
• Final set of 983,734 compounds

22
Docking

• Docking was performed using Openeye's fred
• We considered 4 scoring functions
• Chemgauss3
• OEChemScore
• Shapegauss
• PLP
• The receptor site is specified by hand
• Very fast 4 hours for 10,000 molecules

23
Current Status

• All the drug-like molecules have been docked to
  1YC4
• Other targets are being processed
• Database is being populated
• Currently has 20,000 compounds
• Database is accessible via
• the web ( http//rguha.ath.cx/rguha/dock )
• web services

24
(No Transcript)