Sage Bionetworks a Medical Research Organization to: Create a commons open access bioplatform for bu

1
Sage Bionetworks a Medical Research
Organization toCreate a commons/ open access
bioplatformfor building dynamic disease
modelsConnect scientists as contributorsHarnes
s Emerging Genomics Data to Impact Human
HealthSALSSAugust 21st, 2009
2
.
Models of Disease are not absolute but dynamic
Pre 19th C
Pre 20th C
Pre 21st C
Exist through available technologies
3
Traditional Methods used by Biologists and
Clinicians
Imaging
Blood Tests
Pathology
Physical Exam
Symptoms based Classifications of Disease
4
New Tools- New Understanding- Paradigms Fall
5
Massively Powerful New Tools in last Twenty Years
10,000s of REPORTERS
Attempting to migrate from symptoms/ cellular
pathological to a molecular / personalized basis
of disease
6
Unprecedented Scale of Data Accumulating
7
Reality of Complex Behaviors Manifested by
Living Systems

• The truth is we have little idea on the
  underlying causes of common human diseases.

• We need to more fully embrace the complexity to
  develop a better understanding.

8
Potential Contour of Molecular/Causal Disease
Models
Non-coding RNA network
Altered Neuronetworks
BRAIN
HEART
ENVIRONMENT
GI TRACT
protein network
Diabetes
KIDNEY
metabolite network
ENVIRONMENT
IMMUNE SYSTEM
VASCULATURE
Resistance to Malaria
transcriptional network
ENVIRONMENT
9
Who now uses such Models and Structures in the
Sciences?
10
Our focus has been on generation, assembly, and
integration of data to build models that predict
complex system behavior

• Generate data need to build networks
• Assemble other available data useful for building
  networks
• Integrate and build models
• Test predictions
• Develop treatments
• Design Predictive Markers

Resources gt150M
11
Selection of Targets Using Integrative Genomics
Strategy
Prerequisites for Causal Network Molecular
Disease Models

• Step 1
• Trait-mRNA correlations
• Step 2
• Overlapping QTLs
• Step 3
• Causality test

Clinical Phenotype Adiposity
Trait-RNA Correlation
cQTL
eQTL
Reactive Causal Independent
mRNA expression Gene
Genotype Chr 6, Chr 9
Power of Using Natural variations as Perturbagens
Schadt EE Nat Genet (2005) 205370
12
Preliminary Probabalistic Models- Rosetta /Schadt
Networks facilitate direct identification of
genes that are causal for disease Evolutionarily
tolerated weak spots
Nat Genet (2005) 205370
13
Building networks over multiple tissues
simultaneously

• Expression data from multiple tissues combined to
  build networks that reflect interactions within
  and between tissues
• Red and green nodes to left represent
  hypothalamus and adipose nodes
• We show connections between nodes in this plot
  only if the expression traits are associated
  between tissues

Sweden is a world leader in gathering such data
( ex Johan Bjorkegren/ Jesper Tegner)
14
Differentially connected genes representing gain
of copy numbers in liver tumors are enriched for
genes that predict survival

• Example Brown tumor module
• Enriched in genes undergoing amplification
  (plt0.0073)
• Enriched in genes predicting survival (plt3.8e-7)
  and tumor stage (plt5e-69)
• More highly connected genes are more predictive
  ie network architecture is linked to clinical
  outcome (plt1.15e-12)
• Contains genes highly expressed in human liver
  (plt6.46e-92) and various metabolic pathways
  including amino acid (plt8.67e-31) and lipid
  metabolism (plt8.31e-28) and is possibly centered
  on the mitochondrion (plt2.25e-19)

(John Lamb Eric Schadt John Luk unpublished)
15
Finding Dependent Parts using Probabalistic
Network Disease Models
Macrophage-enriched module
Causal for Diabetes and Obesity Traits

• Test all sub-networks over the entire system to
  identify those that are causal for disease
• One sub-network in particular stood out
• 75 of genes supported as causal for athero
  lesions in this cross (117 of 157 genes) fall in
  this sub-network (p 5.6E-103)
• Near 50 of genes supported as causal for obesity
  (366 of 735 genes) are in this sub-network (p
  3.78E-235)
• Near 45 of genes supported as causal for
  diabetes traits (263 of 571 genes) are in this
  sub-network (p 1.5E-133)
• Sub-network contains hypertension genes
• All enrichments seen across sexes and multiple
  tissues

Angiogenesis-enriched Module
Vascularization-enriched Module
Cell Cycle Module
Nature 452436-444 (2008) Nature 452429-435
(2008)
16
Screening compounds against networks not
targetsallows identification of disease
modification
17
Our ability to integrate compound data into our
network analyses
db/db mouse (p10E(-30))
db/db mouse (p10E(-20) p10E(-100))
ROSI in db/db mouse
18
New models able to predict complex Human
bionetworksrequire coherent data, compatible
tools and models
Harvard Brain Bank (N gt 1000)
Fox Hollow (N gt 1000)
MGH Collab with Lee Kaplan (N gt 1000)
Decode (N gt 1000)
Liver Consortium (N1000)
Visual Cortex
Hippocampus
Pre-Frontal Cortex
Liver
Cerebellum
Adipose
Plaque from Carotid and Peripheral Lesions
Blood
Liver
Adipose
Brain
Gut
trait pairs to consider
These data sets are massive
Isolate RNA and Profile
Isolate DNA
trait pairs
data points
Genotype
tests (26 quadrillion)
19
Transition from building within Rosetta to
Founding a Platform/Commons for representing
Disease Biology
Recognition that the benefits of bionetwork
based molecular models of diseases are great but
that they require significant resources Appreciat
ion that it will require decades of evolving
representations Watching the growth in public
access data on disease biology now possible to
anticipate a transition of disease biology to a
precompetitive space Merck saw that by donating
data, tools, know-how, it might enable an open
access platform / commons to come out of an
incubator phase allowing a potential long term
gain to the whole community provided by evolving
models of disease built via a contributor network
20
Sage Transitioning from Clinicians as Archivists
to a Contributor Community by jointly building
evolving Models of Health and Disease
Current Approaches
Integrative genomic approaches Using causal
bionetwork approaches To build evolving disease
models
Nature Science PLoS
GWAS
N1 Model
TCGA
Requires Reliable Data- Quality Controls Well
annotated- engage standards Curation of
platforms Coordination of Computational
Models Tools for Interactive Work
place (leverage existing public, private efforts
EBI, NCBI etc.)
Linear files Genome Info/Clinical Storing /
Binning Redoing previous experiments
21
Why a Commons Why Sage?
The Commons around Castlemorton are very special.
There are not many places in this county where
you can still see active common grazing and as
such it is hard to understate the cultural and
wildlife value of this process. Grazing of the
commons over centuries has made them what they
are today and is the core tool that delivers the
aims of management on the common. Without the
livestock the Commons would be much poorer place
for the wildlife, and for people such as walkers
or horse riders (due to scrub encroachment
preventing access), and the cultural value of
local graziers turning out stock on the common
would be lost. The grazing is carried out by
local farmers with a legal right to graze on the
Common. In acknowledgment of the valuable
wildlife present on Castlemorton Common, some
areas have been designated by Natural England as
Sites of Special Scientific Interest, a national
designation which gives extra protection to the
common.
22
Sage
To create a commons where integrative bionetworks
evolved by contributor scientists accelerate the
elimination of human diseases.
Vision
Information connectivity
Distributed innovation
Network models of disease
Data, Models, and Tools, with interconnectivity,
data standards, governance rules In a public/
private interface supported by Foundations,
Universities, Government
23
Three Components needed for understanding disease
biology and sharing data in contributor based
world

• Big databases, organized (connected) to
  facilitate integration and model building
• Intel, IBM, Microsoft, Amazon, and so on
• Data integration and construction of predictive
  models
• Computational, math/stat, high-performance
  computing, and biological expertise
• Significant high-performance computing resources
• Tools and educational resources to
• translate complex material
• to a hierarchy of users

24
End state vision open-access contributor network
DNA variation
Assembly of coherent biomedical data
into Probabilistic bionetworks provides a
framework for representing Disease models
Clinical (including EMR)
Intermediate traits (molecular and organismic
physiology)
Network models (living/evolving)
Platform enables model refinement through
microqueries of the Probabilistic bionetworks by
distributed scientists
Contributor scientists
25
Goals for Sage Bionetworks Incubator Phase
Sage Repository
Sage Commons System
Probabilistic Causal Network Models Of Disease
Sage Rules And Governance
26
Example of a Sage Probabilistic Causal Disease
Model for Modulating Cholesterol
27
Sage Organization and Advisors

• Co-Founders
• Stephen H Friend MD PhD
• Eric Schadt PhD
• - Core team members
• Stephen H. Friend MD PhD
• Ex Co-Leader of The Seattle Project, Fred
  Hutchinson Cancer Research Center
• 1994-1998
• Ex Senior VP Merck Co. Inc. Head of Oncology
  Franchise (4/01- 3/09)
• President of Rosetta Inpharmatics (1997-2009)
• Key Scientists in Genetics Group at Rosetta (Comp
  Bio., Stat Gen, Database, Sys. Biol.)
• Directors and Advisors
• Leland Hartwell- President, Fred Hutchinson
  Research Cancer Center
• Richard Lifton MD- Chief, Human Genetics, Yale
  University Med School
• Eric Schadt PhD- CSO Pacific Biosciences
• Dietrich Stephan PhD - CEO Founder Navigenics
• Hans Wigzell MD PhD- Director Emeritus,
  Karolinska Institute

28
Funding Sources for Sage
Government Sources Foundations Biotech/ Pharma
Partners Information Technology
Partners Software Tool Providers
29
Essential Roles by which to enable Sage
Enable specific novel models NHLBI, CHDI-
Huntingtons Fund dataset integration UK
Investigators /EBI , NCBI Provide Support of
Communities wanting to join China, RIKEN, NIH
Intramural Focus us on particular interests
Predictive Vaccinology, NCI-ICSB Join drafting
of the Commons Governance Rules Congress Fall
2009 Designate disease biology data flow into
the public Domain Advocate in Public Request
grantees place datasets into the Commons Fund
development of Tools needed for this to be
scalable
www.sagebase.org
30
Models of Disease are not absolute but dynamic
NeuroNetworks
Diabetes
Exist through available technologies
Resistance to Malaria
www.sagebase.org