SIAM San DiegoT Id kTrey IdekerJuly 11 2008y

Network Biology: Mapping pathways toNetwork Biology: Mapping pathways to understand and diagnose disease

Many kinds of interaction technologiesPHYSICAL GENETIC

Protein-gene(transcriptional, ChIP-Chip22, 39) Epistatic orderings

a < b OR b < aORDERED

Cause and effectSignal transducing

Protein-RNA (RIP-chip80)

Protein-protein(kinase substrate

a b OR b a(EMAP28, 83)

Knock-down expression profiles(RNAi32 deletion mutants36, 37)Signal transducing (kinase-substrate

arrays21, LUMIER81)

Protein-compound82

(RNAi32, deletion mutants36, 37)

Expression QTLs41, 42

UNORDERED

Ambiguous

Protein-protein(co-IP/MS/MS18-20, Y2H15, 84-86)

Synthetic lethalityab << a, b, wt

(SGA88 dSLAM31 71 EMAP28 83gdirectionality Gene-gene (co-regulon87) (SGA88, dSLAM31, 71, EMAP28, 83,

chemogenomic profiling89)

Beyer, Bandyopadhyay, and Ideker Nat. Rev. Genetics (2007)

Like sequence, molecular interaction d t idl iTwo overriding aims:data are rapidly growing…Two overriding aims:

IntAct Database Growthtotal interactions (thousands)

EMBL Database Growthtotal nucleotides (gigabases)

192 1001) Assemble many interactions and types into unified models

96 502) Get rid of false and non-functional interactions

1980 200019900

2008 2003 20062005 20082004 20070

These aims lead to many subproblemsThese aims lead to many subproblems

Mapping transcriptional networks

Networks to interpret genetic variations

Networks to interpret combinatorialNetworks to interpret combinatorial perturbations (e.g. synthetic lethals)

Network evolution

Network-based diagnosisNetwork based diagnosis

Assembly ofAssembly ofphysical and genetic interactions

to map transcriptional circuits

(Chris Workman, Craig Mak with L S Ri h d K l d )Leona Samson, Richard Kolodner)

Can this apparent paradox beCan this apparent paradox be explained by a physical model of the DNA damage response?the DNA damage response?

ChIP-chip measurement of protein→DNA interactions

From Figure 1 of Simon et al. Cell 2001

Mapping DNA Damage Response Networks

Numbers of promoters bound by each of 30 transcription factors (TFs) before and after exposure to methyl-methane sulfonate (MMS)

MMS only−MMS only

+MMS only

bothboth

Workman, Mak, et al. Science 2006

Integration of cause-and-effect interactions with physical networks

Perturbationeffects

TF-promoter bindingPerturbation causes up-regulationProtein-protein bindingPerturbation causes down-regulation

Yeang, Mak et al. Genome Biology 2005

Such methods can yield large regulatory networks

Workman, Mak, et al. Science 2006

Sensitivity of the TF knockout phenotype correlates with its number of regulated targetscorrelates with its number of regulated targets

sed

gen

esre

gula

tem

ber o

f N

u

MMS sensitivity of TF knockout

Mapping Pathways inSynthetic Lethal Networks

(Ryan Kelley, Sourav Bandyopadhyaywith Nevan Krogan)

Finding physical pathways to explain genetic interactions

Genetic Interactions:

• Classical method used to h i d lmap pathways in model

species

• Highly analogous to• Highly analogous tomulti-genic interaction in human disease and combination therapy

• Thousands are being uncovered through systematic studiessystematic studies

Thus as with other types, the number of known genetic

Adapted from Tong et al., Science 2001

number of known genetic interactions isexponentially increasing…

Integration of genetic and physical interactions

160 between-pathway models

101 within-pathway modelsp y

Num interactions:1,102 genetic933 physical933 physical

Kelley and Ideker Nature Biotechnology (2005)

Systematic identification of “parallel pathway” relationships in yeastparallel pathway relationships in yeast

Global organization of genetic linkages between physical modules (A Z)between physical modules (A-Z)

BridgingBridging genetic interactions

Overlapping pathwaysp y

Towards a generative module mapU ti d l f ll hi h id k d l• Use generative model of cell which considers k modules

simultaneously along with their inter-module functional relationships.

• Consider both positive and negative quantitative genetic interactions (alleviating and aggravating)

•

Physical interaction score

Genetic interaction score

Functional maps of protein complexes

with Nevan Krogan

Using protein networks to understand molecular evolution

(with Roded Sharan, Richard Karp, and others)p, )

Cross-comparison of networks:(1) Conserved regions in the presence vs. absence of stimulus(2) Conserved regions across different species

Kelley et al. PNAS 2003Ideker & Sharan Gen Res 2008

Sharan et al. RECOMB 2004Scott et al. RECOMB 2005Sharan & Ideker Nat. Biotech. 2006

Suthram et al. Nature 2005

Plasmodium: a network apart?

Plasmodium-specificprotein complexes

Conserved Plasmodium / Saccharomyces protein complexes

protein complexes

Suthram et al. Nature 2005La Count et al. Nature 2005

Pairwise alignment of the E. coli protein network versus the indicated species;

Sensitivity comparison of different methods

Flannick et al. Genome Research (2006) [Batzoglou Lab]

1960BIOLOGICAL SEQUENCE COMPARISON

Haussler

1980 19901970

First protein

BIOLOGICAL SEQUENCE COMPARISON

Needleman/ Wunsch

PAM, BLOSUM

matrix

Swiss-Prot, GenBank,

EMBL-BankDayhoff,Jukes/Cantor Taylor,

Lipman,

Haussler, Borodovsky,

ChurchillSmith/

Waterman

First protein sequences by Sanger,

othersBLASTDoolittle

Stormo

Public genome-scale

databasesMathematical d l f

Wunschothers

Hidden Markov Models

A new type of data becomes routinely

Mining for motifs and domains

Scoring via transition

probabilities

Automated pairwise li t

Multiple alignment

models of evolution

Fast dynamic programming li t

routinely available

Analysis of global

properties; information

Database queries are

staple of molecular

alignment alignmentalignment information content

Interaction detection with

2-hybrid

MaWishInterologs; evolutionary

biology

Scale-free

Alon’s network motifs

????BIND, DIP, MINT, GRID

2-hybrid, mass. spec.

e o ut o a ymodels PathBLAST property;

robustness

ot s

Sharan/Karp/Ideker

????Ogata/ Kanehisa

BIOLOGICAL NETWORK COMPARISON1990 2005 2010?2001 20032002 2004

Sharan and Ideker Nat. Biotech (2006)

Using protein networks for disease classification

(Han Yu Chuang)

Such methods can yield large regulatory networks

Workman, Mak, et al. Science 2006

Using protein networks to diagnose breast cancer metastasis

Examples of “i f ti“informative

subnetworks”

Network markers are more reproducible and increase classification accuracy of breast cancer metastasisy

Chuang et al. MSB, 2007

www.cytoscape.orgOPEN SOURCE Java platform for integration of systems biology data

•Layout and query of interactionLayout and query of interaction networks (physical and genetic)

•Visual and programmatic integration of molecular state data (attributes)o o ecu a state data (att butes)

•The ultimate goal is to provide the tools to facilitate all aspects of pathway assembly and annotation.p y y

RECENT NEWS

•Version 2 5 released Summer 2007;Version 2.5 released Summer 2007; Scalability+efficiency now equivalent to best commercial packages•The Cytoscape Consortium is a 501(c)3 non-for-profit in the State of Californiay p ( ) p

•The Cytoscape ® Registered Trademark awarded

JOINTLY CODED with Agilent, ISB, Pasteur, Sloan-Kettering, UCSF, Unilever, U Toronto

Genetic Interactions:

Ryan Kelley

DNA Damage Networks

Chris Workman Ryan Kelley,Sourav Bandyopadhyay,Nev Krogan (UCSF)

Craig MakLeona Samson (MIT)Tom Begley (U Albany)

Network Evolution:

Silpa SuthramRoded Sharan (Tel Aviv)

Cancer Diagnosis:Interpretation of eQTLs:

Roded Sharan (Tel Aviv)Richard Karp (Berkeley)

Han Yu Chuang,Steve Briggs,Tom Kipps,E j L (KAIST)

Interpretation of eQTLs:

Silpa SuthramAndreas BeyerYonina Eldar (Technion) Eunjun Lee (KAIST),

Doheon Lee (KAIST)Yonina Eldar (Technion)Richard Karp (Berkeley)

Funding: NIEHS, NIGMS, NSF, Packard, Agilent, Unilever

Websites: www.pathblast.org; www.cytoscape.org

Networks perturbed by individual genetic variations

(Silpa Suthram)

eQTLs generate cause-effect interactionseQTLs generate cause effect interactions

• Expression Quantitative pTrait Loci (eQTLs) look for associations between SNPs and geneSNPs and gene expression levels.

• All-vs-All analysis: ALL SNP l t d f G

enes

SNPs are evaluated for association with ALL gene expression levels.

G• This process can

generate thousands of associations SNPsassociations. SNPs

Cause and effect interactionsCause and effect interactions

Knock-down expression profiles(RNAi deletion mutants)(RNAi, deletion mutants)

ORORExpression QTLs

Knockout causes up-regulationKnockout causes down-regulation

Integration of cause-and-effect interactions with physical networks

Perturbationeffects

TF-promoter bindingPerturbation causes up-regulationProtein-protein bindingPerturbation causes down-regulation

Yeang, Mak et al. Genome Biology 2005

ExamplesExamples

Suthram et al. Nature/EMBO MSB 2008Silpa Suthram