A community-driven annotation platform for structural genomics

advertisement
A community-driven annotation platform for
structural genomics
Adam Godzik
and the JCSG Bioinformatics Team
Workshop on the Biological Annotation of Novel Proteins, March 7-8, 2008
Biomedical theme: Central Machinery of Life -proteins conserved in all kingdoms of life
Biological theme: Complete coverage of Thermotoga maritima
Science is all about communication
• Since late XIX century, a dominant way of communicating scientific
results is through peer-reviewed manuscripts
• Pro
– Peer review ensures quality
– Enforces a “publishable unit” – decreases noise in the
“communication space”
– Authorship rules ensure proper distribution of credit in a
system that is well integrated with system of promotions
and evaluations
• Con
– Significant time lag and additional costs
– Enforces a “publishable unit” – below the threshold results
are lost
– Not scalable with high throughput data production
Increasingly, it’s not the only game in
town
• Databases and automated annotation protocols
– pro: fast, machine searchable, scalable
– con: difficult to ensure quality and assign credit, put the
burden of expertise on the user
• Wikipedia
– pro: harnesses power of community, scalable
– con: unreliable, difficult to ensure quality and assign
credit
Can we have the best of all worlds?
Peer-reviewed
manuscript
Fast, accurate,
scalable
Wikipedia
entry
Automated
database annotation
Structure determination in PSI centers is done on a
semi-automated assembly line
target
selection
expression
cloning
imaging
harvesting
xtal mounting
xtal screening
annotation
publication
PDB
•
•
•
•
purification
crystallization
data collection
struc. validation
phasing
tracing
struc. refinement
Joint Center for Structural Genomics
One of four large scale (production) centers of PSI2
~ 600 structures deposited in the PDB
Sustained pace of ~15 PDB depositions per month
…and the pace of structure determination far outstrips
the pace of our publications…
PSI-1
PSI-2
Structure collage: PSI
Publication statistics : http://olenka.med.virginia.edu/psi/
Why?
• Speed: 2-3 months from target selection to
structure, 2-3 structures per week in each center
• Assembly line process, no time to develop
“special relationship” with each protein
• Structures are not associated with ongoing
biochemical and biological research.
• Targets selected based on novelty, no expertise
available anywhere, difficult to reach “publishable
unit”
We are not alone …
• Bacterial genomes
– 1995-2000 - every new sequenced genome led to a
Nature/Science publication
– with 500+ genomes, an increasing percentage of them
never become a single focus of a specific publication
– Community based annotation efforts become the best
source of information (SEED)
WEB 2.0 is reshaping how we share
information:
Wikipedia
Citizendium
Scholarpedia
Google Knols
OpenWetware
Wikiomics
WikiProteins
TOPSAN
Communities of globally distributed peers (Networks)
built around rich, collaborative environments.
How can we tap into an ultimate
research tool?
• Search engines are
becoming serious
research tools
• Google indexes research
papers, books, wikipedia
pages
• Semi-natural language
searches
Structural coverage of many genomes (here
T.maritima) approaches completeness
80
70
% covered
60
~73% of feasible targets
in PDB
sequence identity > 30%
blast e-value < 0.001
FFAS score < -9.5
50
40
30
20
10
0
1980
1985
1990
1995
year
2000
2007
Which brings attention from
broader communities
http://research.calit2.net/metagenomics/thermotoga
We can utilize other information
•
•
•
Metabolic reconstruction of
T.maritima was done in collaboration
with UCSD Systems Biology Lab
(Bernard Palsson)
Model is consistent with all the
published experimental data on TM
(see Ines Thiele poster)
First generation model covers 479
genes (1398 are not in the model),
492 metabolites
–
–
–
Proteins coded by 113 of these
genes have been solved (71 at
JCSG, 28 at other PSI centers)
320 have be modeled
We know at least approximate
structure of ALL the proteins in the
reconstruction
And bring it together to help make sense of the
structures and see them in the full context
All available information about a protein on one
page
We try to combine automated, database driven annotations
with expert curated input.
Annotation:
• Feeds from public databases
• Expert-curated information
Content management:
• Wiki-style editing (WYSIWYG editor)
• Page-level access control
• Structured fields + free text
• Instant publication
• Always open for comments and edits
Quality control & authorship:
• Encourage community collaboration
• JCSG scientists & invited peers
• Many authors - no contribution too
small
• Lead authors (editors) in charge of
releases
TOPSAN content:
Protein Groups
Browsing Options
JCSG: Structures / Structure Notes / TOPSAN
278 of 593 structures have an annotation on TOPSAN
Members of the biological community can utilize
PSI structures only when they are aware of them
Functionally well characterized
enzyme and is also a new fold.
PDB ID: 3C8W
TargetDB: 376561
TOPSAN access statistics- Jan to Mar 08
1143 visits from
rest of the world.
Google sent us the largest number of visitors.
GNF & TSRI
Crystallomics Core
Scott Lesley
Mark Knuth
Heath Klock
Marc Deller
Dennis Carlton
Polat Abdubek
Sanjay Agarwalla
Connie Chen
Thomas Clayton
Dustin Ernst
Julie Feuerhelm
Regina Gorski
Anna Grzechnik
Joanna C. Hale
Thamara Janaratne
Hope Johnson
Sachin Kale
Daniel McMullan
Edward Nigoghossian
Amanda Nopakun
Linda Okach
Jessica Paulsen
Christina Puckett
Sebastian Sudek
Jessica Canseco
UCSD &
Burnham
Bioinformatics Core
John Wooley
Adam Godzik
Lukasz Jaroszewski
Slawomir Grzechnik
Sri Krishna Subramanian
Andrew Morse
Tamara Astakhova
Lian Duan
Piotr Kozbial
Dana Weekes
Natasha Sefcovic
Josie Alaoen
TSRI
NMR Core
Kurt Wüthrich
Reto Horst
Margaret Johnson
Amaranth Chatterjee
Michael Geralt
Wojtek Augustyniak
Jin-Kyu Rhee
Biswaranjan Mohanty
Bill Pedrini
Pedro Serrano
Stanford /SSRL
Structure
Determination Core
Keith Hodgson
Ashley Deacon
Mitchell Miller
Hsiu-Ju (Jessica) Chiu
Debanu Das
Kevin Jin
Abhinav Kumar
Winnie Lam
Silvya Oommachen
Christopher Rife
Scott Talafuse
Christine Trame
Qingping Xu
Henry van den Bedem
Ronald Reyes
TSRI
Administrative Core
Ian Wilson
Marc Elsliger
Gye Won Han
David Marciano
Henry Tien
Xiaoping Dai
Lisa van Veen
Scientific Advisory Board
Sir Tom Blundell
Univ. Cambridge
Homme Hellinga
Duke University Medical Center
James Naismith
The Scottish Structural Proteomics facility
Univ. St. Andrews
James Paulson
Consortium for Functional Glycomics,
The Scripps Research Institute
Robert Stroud
Center for Structure of Membrane Proteins,
Membrane Protein Expression Center
UC San Francisco
Soichi Wakatsuki
Photon Factory, KEK, Japan
James Wells
UC San Francisco
Todd Yeates
UCLA-DOE, Inst. for
Genomics and Proteomics
The JCSG is supported by the NIH Protein
Structure Initiative grant U54 GM074898
from the National Institute of General
Medical Sciences (www.nigms.nih.gov).
Thermotoga browser acknowledgments
•
•
•
•
•
•
•
•
•
Co-PI of the project - Andrei Osterman (the biochemistry side, specific examples)
The JCSG team - for all the structures, focus on Thermotoga and CML
Bernard Palsson group and Ines Thiele for work with Thermotoga reconstruction and
model simulations
The JCMM team for structure modeling
Krzysztof Ginalski and bioinfor server team for assistance with “borderline” predictions
Ying Zhang (JCMM) - finalizing the metabolic reconstruction, network and fold
distribution analysis
Dana Weekes (JCSG) - first pass on the Thermotoga metabolic reconstruction, TM
TOPSAN pages
Craig Shepherd (JCMM) - network visualization
Zhanwen Li (JCMM) - modeling and fold assignments
Download