Transcript Dialogue DataGrid Motivating Applications

Dialogue DataGrid
Motivating Applications
Joel Saltz, MD, PhD
Chair Biomedical Informatics
College of Medicine
The Ohio State University
Dialogue DataGrid
• Relational databases, files, XML databases,
object stores
• Strongly typed
• Multi-tiered metadata management system
• Incorporates elements from OGSA-DAI,
Mobius, caGrid, STORM, DataCutter, GT4 …
• Scales to very large data, high end
platforms
Requirements
• Support or interoperate with caGrid,
eScience infrastructure
• Interoperate with or replace SRB
• Well defined relationship to Globus Alliance
• Services to support high end large scale
data applications
• Design should include semantic metadata
management
• Well thought out relationship to commercial
products (e.g. Information Integrator,
Oracle)
Motivating Application Class I: Phenotype
characterization
• Information heterogeneity,
data coordination and data
size
• Synthesize information from
many high throughput
information sources
• Sources can include multiple
types of high throughput
molecular data and multiple
imaging modalities.
• Coordinated efforts at multiple
sites
• Detailed understanding of
biomedical phenomena will
increasingly involve the need
to analyze very large high
resolution spatio-temporal
datasets
Structural Complexity
Example Questions (Phenotypes
associated with Rb knockouts)
1.
What are the
mechanisms of fetal
death in mutant mice?
2.
What structural
changes occur in the
placenta?
3.
How different are the
structural changes
between the wild and
mutant types?
4.
…
Rb+
Rb-
Dataset Size: Systems Biology
Future big science animal
experiments on cancer,
heart disease, pathogen
host response
Basic small mouse is 3 cm3
1 μ resolution – very roughly
1013 bytes/mouse
Molecular data (spatial
location) multiply by 102
Vary genetic composition,
environmental
manipulation, systematic
mechanisms for varying
genetic expression; multiply
by 103
Total: 1018 bytes per big
science animal
experiment
Now: Virtual Slides
(roughly 25TB/cm2 tissue)
Compare phenotypes of normal vs Rb
deficient mice
Slides/Slices
Alignment
Placenta
Visualization
Segmentation
Computational Phenotyping
Challenges
• Very large datasets
• Automated image analysis
• Three dimensional reconstruction
• Motion
• Integration of multiple data sources
• Data indexing and retrieval
Large Scale Data Middleware Requirements
• Spatio-temporal datasets
• Very large datasets
– Tens of gigabytes to 100+ TB data
• Lots of datasets
– Up to thousands of runs for a study are possible
• Data can be stored in distributed collection
of files
• Distributed datasets
– Data may be captured at multiple locations by multiple
groups
– Simulations are carried out at multiple sites
• Common operations: subsetting, filtering,
interpolations, projections, comparisons,
frequency counts
Very Large Dataset Hardware is Proliferating
LinTel boxes (PvFS/
Active Disk Archive) (20)






D V D

D V D
(2)
890 MB/s through
MetaData Servers
(2)

D V D

D V D
(2)
(2)
890 M
B/s Th
rough
put

D V D
(2)
)
(2
(2)
D V D
DVD
DVD
DVD
DVD





DVD
DVD
DVD
DVD
DVD





DVD
DVD
DVD
DVD
DVD





DVD
DVD
DVD
DVD
DVD
(40 - 2 per xSeries)
10 GB/s
)
(2

DVD
(40 - 2 per T600)
384 MB/s throughput
put
r)
Cisco Directors 9509
ve ut
er hp
r s oug
e
p thr
4
(4)
6 B/s MB/s throughput
(1 M772
0
(4)
89
(4)
772 MB/s throughput
FAStT600 Turbo (20)
Scratch / Archive Storage Pool (310/420 TB)
(4)
772 MB/s throughput
(4)
772 MB/s throughput
SAN Volume Controller
(4 servers)
FAStT900 (4)
Core Storage Pool (35/50 TB) with SAN.FS
Backup Storage
3584 Tape
1 L32 2 D32
Actual: 640 cartridges @ 200
GB for a total of 128 TB
4 drives
max drive data rate is 35 MB/s
• 50 TB of performance
storage
– home directories, project
storage space, and longterm frequently accessed
files.
• 420 TB of
performance/capacity
storage
– Active Disk Cache compute jobs that require
directly connected storage
– parallel file systems, and
scratch space.
– Large temporary holding
area
• 128 TB tape library
– Backups and long-term
"offline" storage
Our Example: Ohio Supercomputing Center Mass Storage
Testbed
STORM Services
• Query
• Meta-data
• Indexing
• Data Source
• Filtering
• Partition
Generation
• Data Mover
STORM Results
Seismic Datasets
10-25GB per file.
About 30-35TB of Data.
STORM I/O Performance
4500
4000
Bandwidth (MB/s)
3500
3000
2 Threads
2500
4 Threads
2000
Max
1500
1000
500
0
1
2
4
# XIO nodes
8
16
Motivating Application II: caBIG In vivo
Imaging Workspace
Testbed
• Study the effects of image acquisition and reconstruction
parameters (i.e. slice thickness, reconstruction kernel and
dose) on CAD and on human ROC.
– use multiple datasets and several CAD algorithms to
investigate the relationship between radiation dose and
nodule detection ROC.
• Cooperative Group Imaging Study Support
– Children’s Oncology Group: quantify whether perfusion
study results add any additional predictive value to the
current battery of histopathological and molecular studies
– CALGB: Grid based analysis of PET/CT data to support
phase I, II studies
– NTROI: Grid based OMIRAD -- registration, fusion and
analysis of MR and Diffusive Optical Tomography (DOT).
CAD Testbed Project RSNA 2005 (joint with
Eliot Siegel et al at Univ. Maryland)
• Expose algorithms and data management as Grid Services
• Remote execution of multiple CAD algorithms using
multiple image databases
• CAD algorithm validation with larger set of images
• Better research support — recruit from multiple institutions,
demographic relationships, outcome management etc.
• Remote CAD execution - reduced data transfer & avoid
need to transmit PHI
• CAD compute farms that reduce the turnover time
• Scalable and open source — caBIG standards
Architecture
Image Data Service
•Expose data in DICOM PACS with grid service
wrappers
•An open source DICOM server — Pixelmed
•XML based data transfer
5 Participating Data Services
3x Chicago
1x Columbus
1x Los Angeles
CAD Algorithm Service
• Grid services for algorithm invocation and image retrieval service
• caGRID middleware to wrap CAD applications with grid services
• Report results to a result storage service
caGrid Introduce
facilitates service
creation
GUMS/CAMS is used
to provide secure data
communication and
service invocation
CAD algorithms provided by iCAD and Siemens Medical Solutions.
Prototypes for investigational use only; not commercially available
Framework Support Services
• Result storage server — A distributed XML database for caching
CAD results
• GME — Manage communication and data protocols
User Interface
Available
data
services
DICOM
image viewer
17
5
14
12
Queried
results
18
15
16
Slice = 127
W/L = 2200/-500
Click to browse images, submit
CAD analysis, and view results
Motivating Application III: Integrative
Cancer Biology Center on Epigenetics (PI
Tim Huang, OSU)
• TGFβ/Smad4 targets are regulated via
epigenetic mechanisms. Promoter
hypermethylation is a hallmark of this
epigenetically mediated silencing. In
this study, we have combined both
chromatin immunoprecipitation
microarray (ChIP-chip) and differential
methylation hybridization (DMH) to
investigate this epigenetic mechanism
in ovarian cancer
Translating a goal into workflow
n9
n1
n8
K
Data Mining
(ex: clustering
Literature Survey
- Experimentally verified
TGF-B target genes
- Housekeeping Genes
Clinical data
Data Collection-Genome
-UCSC Genome
-BLAT alignment
Analytical service J
Data source K
n10
Construct KbTSMAD
(Knowledgebase of the
TGF-B/SMAD signaling pathway)
L
n7
Normilization
with statistical tools
Experimental Results
From Other Groups
A
Analytical service I
KbTSMAD
Data source L
Data source A
n2
n3
B
C
Description of
Experiment
Chip-chip Results
(Microarray data)
D
E
Description of
Experiment
DMH Results
(Microarray
Data)
DMH Experiment
Chip-on-chip Experiment
Data source B and C
Data source D and E
n5
n4
Chip design
G
Analytical service F
Custom chip design info
(e.g. from Agilent)
Data source G
n6
H
Candidate cutting Enzyme
Information
Data source H
ArrayAnnotator
Application of caGrid to the
workflow
• Application needs to support access to a diverse set of
databases and execution of different analysis methods
• Data services
–
–
–
–
–
–
KbSMAD
Chip information from chip company
Enzyme data
Clinical data
Experimental results
Experimental design
• Analytical services
– Designing a custom array
– Normalization
– Data mining (ex: clustering)
Example: Prototype of Clone
Annotation Analytical Service
• Analytical Service: ArrayAnnotator
 Goal: Provide a annotation for each clone to select a subset of clones
among 400,000 candidate clones to design a custom array for DMH
experiment
 Clone selection criteria




Clones within a promoter region
Clones with proper internal and external cuts
Clones within CpG island region and/or high CG contents
Clones with Transcription Factor binding sites
 Input: CloneType information
 extended sequence, enzyme info, genomic location, etc
 Functions
• Determine external cut locations around a clone region (e.g., cut-site by BfaI)
• Examine the internal cut around a clone region (e.g., cut-site by HapII, HinpII,
and MCrBc)
• Identify the location of clone in genome
• Show ether it is within promoter region or not
• Calculate CG content and overlapping with CpG islands
• Identify which Transcription Factor binding sites are overlapped with clones
Example caGrid Usage in P50 chip design
application
Query (geneId)
Result: List of clones
Clone Info
Data Services
1
2
Genome Sequence
Data Source
Result: extended genome sequence
4 of clone
3 Query
Annotation
Analytical Service
5
Request (cloneInfo)
6
Chip design application
Result: annotation (CpG, cutsite,
promoter region, etc)
ArrayAnnotator output (Hao Sun, Ramana
Davuluri)
Multiscale Laboratory Research Group
Ohio State University
Joel Saltz
Gagan Agrawal
Umit Catalyurek
Dan Cowden
Mike Gray
Tahsin Kurc
Shannon Hastings
Steve Langella
Scott Oster
Tony Pan
DK Panda
Srini Parthasarathy
P. Sadayappan
Sivaramakrishnan (K2)
Michael Zhang
The Ohio
Supercomputer Center
Stan Ahalt
Jason Bryan
Dennis Sessanna
Don Stredney
Pete Wycoff
Microscopy Image Analysis
• Biomedical Informatics
– Tony Pan
– Alexandra Gulacy
– Dr. Metin Gurcan
– Dr. Ashish Sharma
– Dr. Kun Huang
– Dr. Joel Saltz
• Computer Science and
Engineering
– Kishore Mosaliganti
– Randall Ridgway
– Richard Sharp
●
Pathology
–Dr.
Dan Cowden
Human Cancer
Genetics
●
–Pamela
Wenzel
–Dr.
Gustavo Leone
–Dr.
Alain deBruin
caGrid Team
• National Cancer
Institute
– Peter Covitz
– Krishnakant
Shanbhag
• Ohio State
University
– Shannon
Hastings
– Tahsin Kurc
– Stephen
Langella
– Scott Oster
– Joel Saltz
• SAIC
–
–
–
–
–
Tara Akhavan
Manav Kher
William Sanchez
Ruowei Wu
Jijin Yan
• Booze | Allen |
Hamilton
– Manisundaram
Arumani
• Panther
Informatics Inc.
– Nick Encina
– Brian Gilman
RSNA 2005 Team
Tony Pan, Stephen Langella, Shannon Hastings, Scott Oster,
Ashish Sharma, Metin Gurcan, Tahsin Kurc, Joel Saltz
Department of Biomedical Informatics
The Ohio State University Medical Center, Columbus OH
Eliot Siegel, Khan M. Siddiqui
University of Maryland School of Medicine, Baltimore, MD
Thanks to Siemens, ICAD for supplying CAD algorithms