stephen friend complex traits: genomics and computational approaches 2012-02-23
DESCRIPTION
Stephen Friend, Feb 23, 2012. Complex Traits: Genomics and Computational Approaches, Breckenridge, COTRANSCRIPT
If the physicists do it, the software engineers do it, Why can’t we do it?:
Moving beyond linear investigations Both of the science and of how we work
Integrating layers of omics data models and building using compute spaces capable of enabling models
to be evolved by teams of teams
Stephen Friend MD PhD
Sage Bionetworks (Non-Profit Organization) Seattle/ Beijing/ Amsterdam
February 23, 2012
So what is the problem?
Most approved therapies were assumed to be monotherapies for diseases represen4ng homogenous popula4ons
Our exis4ng disease models o9en assume pathway knowledge sufficient to infer correct therapies
Familiar but Incomplete
Reality: Overlapping Pathways
The value of appropriate representations/ maps
Equipment capable of generating massive amounts of data
“Data Intensive” Science- Fourth Scientific Paradigm
Open Information System
IT Interoperability
Host evolving computational models in a “Compute Space”
WHY NOT USE “DATA INTENSIVE” SCIENCE
TO BUILD BETTER DISEASE MAPS?
what will it take to understand disease?
DNA RNA PROTEIN (dark maHer)
MOVING BEYOND ALTERED COMPONENT LISTS
2002 Can one build a “causal” model?
Preliminary Probabalistic Models- Rosetta /Schadt
Gene symbol Gene name Variance of OFPM explained by gene expression*
Mouse model
Source
Zfp90 Zinc finger protein 90 68% tg Constructed using BAC transgenics Gas7 Growth arrest specific 7 68% tg Constructed using BAC transgenics Gpx3 Glutathione peroxidase 3 61% tg Provided by Prof. Oleg
Mirochnitchenko (University of Medicine and Dentistry at New Jersey, NJ) [12]
Lactb Lactamase beta 52% tg Constructed using BAC transgenics Me1 Malic enzyme 1 52% ko Naturally occurring KO Gyk Glycerol kinase 46% ko Provided by Dr. Katrina Dipple
(UCLA) [13] Lpl Lipoprotein lipase 46% ko Provided by Dr. Ira Goldberg
(Columbia University, NY) [11] C3ar1 Complement component
3a receptor 1 46% ko Purchased from Deltagen, CA
Tgfbr2 Transforming growth factor beta receptor 2
39% ko Purchased from Deltagen, CA
Networks facilitate direct identification of genes that are
causal for disease Evolutionarily tolerated weak spots
Nat Genet (2005) 205:370
DIVERSE POWERFUL USE OF MODELS AND NETWORKS
50 network papers http://sagebase.org/research/resources.php
List of Influential Papers in Network Modeling
(Eric Schadt)
Equipment capable of generating massive amounts of data A-
“Data Intensive” Science- Fourth Scientific Paradigm Score Card for Medical Sciences
Open Information System D-
IT Interoperability D
Host evolving computational models in a “Compute Space F
.
We still consider much clinical research as if we were “hunter gathers”- not sharing
TENURE FEUDAL STATES
Clinical/genomic data are accessible but minimally usable
Little incentive to annotate and curate data for other scientists to use
Mathematical models of disease are not built to be
reproduced or versioned by others
Lack of standard forms for future rights and consents
Lack of data standards..
Sage Mission
Sage Bionetworks is a non-profit organization with a vision to create a “commons” where integrative bionetworks are evolved by
contributor scientists with a shared vision to accelerate the elimination of human disease
Sagebase.org
Data Repository
Discovery Platform
Building Disease Maps
Commons Pilots
Sage Bionetworks Collaborators
Pharma Partners Merck, Pfizer, Takeda, Astra Zeneca, Amgen, Johnson &Johnson
27
Foundations Kauffman CHDI, Gates Foundation
Government NIH, LSDF, NCI
Academic Levy (Framingham) Rosengren (Lund) Krauss (CHORI)
Federation Ideker, Califano, Nolan, Schadt
A) Miller 159 samples B) Christos 189 samples
C) NKI 295 samples
D) Wang 286 samples
Cell cycle
Pre-mRNA
ECM
Immune response
Blood vessel
E) Super modules
Zhang B et al., Towards a global picture of breast cancer (manuscript).
28
NKI: N Engl J Med. 2002 Dec 19;347(25):1999.
Wang: Lancet. 2005 Feb 19-25;365(9460):671.
Miller: Breast Cancer Res. 2005;7(6):R953.
Christos: J Natl Cancer Inst. 2006 15;98(4):262.
Model of Breast Cancer: Co-expression JUN ZHU
What is this?
Bayesian networks enriched in inflammaQon genes correlated with disease severity in pre-‐frontal cortex of 250 Alzheimer’s paQents.
What does it mean?
InflammaQon in AD is an interacQve mulQ-‐pathway system. More broadly, network structure organizes complex disease effects into coherent sub-‐systems and can prioriQze key genes.
Are you joking?
Gene validaQon shows novel key drivers increase Abeta uptake and decrease neurite length through an ROS burst. (highly relevant to AD pathology)
CHRIS GAITERI-‐ALZHEIMER’S
Elias Chaibub Neto1, Aimee T. Broman2, Mark P. Keller2, Alan D. Attie2, Bin Zhang1, Jun Zhu1, Brian S. Yandell2
1 Sage Bionetworks, Seattle, WA USA; 2 University of Wisconsin-Madison, Madison, WI USA
Causal Model Selection Hypothesis Tests in Systems Genetics
Abstract
Current efforts in systems genetics have focused on the development of statistical approaches aiming to disentangle causal relationships among molecular phenotypes in segregating populations. Model selection criterions, such as the AIC and BIC, have been widely used for this purpose, in spite of being unable to quantify the uncertainty associated with the model selection call. Here we propose three novel hypothesis tests to perform model selection among models representing distinct causal relationships. We focus on models composed of pairs of phenotypes and use their common QTL to determine which phenotype has a causal effect on the other, or whether the phenotypes are not causally related, and are only statistically associated. Our hypothesis tests are fully analytical and avoid the use of computationally expensive permutation or re-sampling strategies. They adapt and extend Vuong's (and Clarke’s) model selection test to the comparison of four possibly misspecified models, handling the full range of possible causal relationships among a pair of phenotypes. We evaluate the performance of our tests against the AIC, BIC and a published causality inference test in simulation studies. Furthermore, we compare the precision of the causal predictions made by the methods using biologically validated causal relationships extracted from a database of 247 knockout experiments in yeast. Overall, our model selection hypothesis tests achieve higher precision than the alternative methods at the expense of reduced statistical power.
Vuong’s Model Selection Test
Vuong's test derives from the Kullback-Leibler Information Criterion (KLIC).
Let h0(y | x) represent the true model.
Consider the parametric family of conditional models: {f(y | x; φ): φ ϵ Ф}.
Then KLIC(h0, f) = E0[log h0(y | x)] – E0[log f(y | x; φ)],
where the expectation E0 is computed w.r.t h0(y, x), and φ* is the parameter value that minimizes KLIC(h0, f).
Consider two models: f1 ≡ f1(y | x; φ1*) and f2 ≡ f2(y | x; φ2*).
Model f1 is a better approximation of h0 than f2 if and only if
KLIC(h0, f1) < KLIC(h0, f2) E0[log f1] > E0[log f2].
Let LR12 = log f1 – log f2. Then we test
H0: E0[LR12] = 0, H1: E0[LR12] > 0, H2: E0[LR12] < 0.
The quantity E0[LR12] is unknown, but the sample mean and variance of
LR�12,i = log f�1,i – log f�2,i, f�1 ≡ f(y | x; φ�1), φ�1 ≡ ML est. of φ1
converve a.s. to E0[LR12] and Var0[LR12] = σ12.12 .
Let LR�12 = ∑ LR�12,i , then under H0
(n σ�12.12 )−1/2 LR�12 →d N(0, 1).
If different models have different dimensions we consider
LR�*12 = LR�12 – D12
where D12 represents a difference of AIC or BIC penalties, and adopt the test statistic
Z12 = (n σ�12.12 )−1/2 LR�*12 .
Clarke’s Model Selection Test
Represents a non-parametric version of Vuong’s test.
Vuong’s null: the mean log-likelihood ratio is 0. Clarke’s null: the median log-likelihood ratio is 0.
Paired sign test on log-likelihood scores:
Scores: (LR�12,1 , LR�12,2 , LR�12,3 , LR�12,4 , LR�12,5 , … , LR�12,n ) Signs: ( + , − , + , + , − , … , + )
Let, T12 = {# of positive signs}. Then under Clarke’s null
T12 ~ Binomial(n, 1/2).
Causal Model Selection Tests (CMST)
In our applications we consider four models: M1, M2, M3 and M4.
We derive intersection-union tests based on six separate Vuong (Clarke) tests:
f1 vs f2 , f1 vs f3 , f1 vs f4 , f2 vs f3 , f2 vs f4 , f3 vs f4
We propose three distinct CMST tests: (1) parametric, (2) non-parametric, and (3) joint-parametric CMST tests.
Parametric CMST:
H0: model M1 is not closer to the true model than M2, M3 or M4. H1: model M1 is closer to the true model than M2, M3 and M4.
H0: { E0[LR12] = 0 } { E0[LR13] = 0 } { E0[LR14] = 0 } H1: { E0[LR12] > 0 } ∩ { E0[LR13] > 0 } ∩ { E0[LR14] > 0 }
The rejection region and p-value for this IU-test are given by:
min{z12 , z13 , z14} > cα , p1 = max{p12 , p13 , p14}.
Non-parametric CMST:
Analogous to the parametric CMST. Just replace Vuong’s by Clarke’s tests.
Joint parametric CMST:
Simple application of Vuong tests, overlooks the dependency among the test statistics.
Let S1 represent the sample covariance matrix of LR�12,i , LR�13,i and LR�14,i.
Under regularity conditions we have that S1 converges a.s. to Σ1.
It follows from the MCT and Slutsky’s theorem that when
( E0[LR12] , E0[LR13] , E0[LR14] )T = ( 0 , 0 , 0 )T
we have that
Z1 = n−1/2 diag(S1)−1/2 LR�1 →d N3(0 , ρ1)
where LR�1 = ( LR�12 , LR�13 , LR�14 )T and ρ1 = diag(S1)−1/2 Σ1 diag(S1)−1/2
We consider the hypotheses
H0: min{ E0[LR12] , E0[LR13] , E0[LR14] } ≤ 0 H1: min{ E0[LR12] , E0[LR13] , E0[LR14] } > 0
and adopt the test statistic W1 = min{Z1}. The p-value is computed as
P(W1 ≥ w1) = P(Z12 ≥ w1 , Z13 ≥ w1 , Z14 ≥ w1).
Simulation Study
We conducted a simulation study generating data from the models on the Figure below.
The results are shown below:
Yeast Data Analysis
We analyzed the yeast genetical genonics data set from Brem and Kruglyak (2005).
We evaluated the precision of the causal predictions made by the methods using validated causal relationships extracted from a data-base of 247 knock-out experiments (Hughes 2000, Zhu 2008).
In total, 46 of the ko-genes showed significant eQTLs, and we tested a total of 4,928 ko-gene/putative target gene relations.
Pairwise Causal Models
Given a pair of phenotypes, Y1 and Y2, that co-map to the same quantitative trait loci, Q, we consider the following models:
Conclusions
Advantages of the Causal Model Selection Tests:
1- Fully analytical hypothesis tests that avoid the use of computationally expensive permutation or re-sampling techniques.
2- Achieve better controlled type I error rates.
3- Achieve higher precision rates.
Main disadvantage: lower statistical power.
ELIAS NETO
Causal Model Selection Hypothesis Tests in Systems Genetics
The Schadt et al. (2005) approach was based on a penalized likelihood model selection approach, were we simply select the model with the best score.
The proposed hypothesis test allows us to attach a p-value to the selected model and, in this way, allows the quantification of the uncertainty associated with the model selection call.
The proposed tests are fully analytical and avoid computationally expensive permutation and re-sampling techniques.
ELIAS NETO
Liver Adipose
FaDy acids
Hypothalamus
Macrophage/ inflamma4on
Lep4n signaling
Phagocytosis-‐ induced lipolysis
Phagocytosis-‐ induced lipolysis
M1 macrophage
A mulQ-‐Qssue immune-‐driven theory of weight loss
ZHI WANG
RULES GOVERN
PLAT
FORM
NEW
MAP
S
PLATFORM Sage Platform and Infrastructure Builders-
( Academic Biotech and Industry IT Partners...)
PILOTS= PROJECTS FOR COMMONS Data Sharing Commons Pilots-
(Federation, CCSB, Inspire2Live....)
Why not share clinical /genomic data and model building in the ways currently used by the software industry (power of tracking workflows and versioning
Leveraging Existing Technologies
Taverna
Addama
tranSMART
Watch What I Do, Not What I Say sage bionetworks synapse project
Reduce, Reuse, Recycle sage bionetworks synapse project
Most of the People You Need to Work with Don’t Work with You
sage bionetworks synapse project
My Other Computer is Cloudera Amazon Google
sage bionetworks synapse project
Sage Metagenomics Project
• > 10k genomic and expression standardized datasets indexed in SCR • Error detection, normalization in mG • Access raw or processed data via download or API in downstream analysis • Building towards open, continuous community curation
Processed Data (S3)
Sage Metagenomics using Amazon Simple Workflow
Full case study at http://aws.amazon.com/swf/testimonials/swfsagebio/
Amazon SWF and Synapse
• Maintains state of analysis • Tracks step execution • Logs workflow history • Dispatches work to Amazon or
remote worker nodes • Efficiently match job size to
hardware • Provides error handling and
recovery
• Hosts raw and processed data for further reuse in public or private projects
• Provides visibility into intermediate results and algorithmic details
• Allows programmatic access to data; integration with R
• Provides standard terminologies for annotations
• Search across data sets
Synapse Roadmap
Q1-2012 Q2-2012 Q3-2012 Q4-2012 Q1-2013 Q2-2013
Synapse Platform Functionality
Data / Analysis Capabilities
Q3-2013 Q4-2013
Internal Alpha Public Beta Testing Synapse 1.0 Synapse 1.5 Future
• Data Repository • Projects and security • R integration • Analysis provenance
• Search • Controlled Vocabularies • Governance of restricted data
• 40+ manually curated clinical studies • 8000 + GEO / Array Express datasets • Clinical, genomic, compound sensitivity • Bioconductor and custom R analysis
• TCGA • METABRIC breast cancer challenge
• Workflow templates • Publishing figures • Wiki & collaboration tools • Integrated management of cloud resources
• Social networking • User-customized dashboards • R Studio integration • Curation tool integration
• Predictive modeling workflows • Automated processing of common genomics platforms
• TBD: Integrations with other visualization and analysis packages
INTEROPERABILITY
INTEROPERABILITY
Genome Pattern CYTOSCAPE tranSMART I2B2
SYNAPSE
Open Network Biology is an open access journal that publishes arQcles relaQng to predicQve, network-‐based models of living systems linked to the corresponding coherent data sets upon which the models are based. In addiQon to arQcles describing these large data sets, the journal also welcomes submissions of original research, sobware and methods, along with reviews and commentary, relevant to the emerging field of network biology.
Submit your manuscript and benefit from: • High visibility for arQcles through unrestricted online access • Free arQcle redistribuQon under a CreaQve Commons aHribuQon license
• No limits on arQcle length, addiQonal files, colour figures or movies
• Rapid, immediate open access publicaQon on acceptance • An integrated repository for network model data and code
Now accep4ng submissions
Editor-‐in-‐Chief Eric Schadt (USA)
www.opennetworkbiology.com
CTCAP Arch2POCM The FederaQon Portable Legal Consent Sage Congress Project
Five Pilots involving Sage Bionetworks
RULES GOVERN
PLAT
FORM
NEW
MAP
S
Clinical Trial Comparator Arm Partnership (CTCAP)
Description: Collate, Annotate, Curate and Host Clinical Trial Data with Genomic Information from the Comparator Arms of Industry and Foundation Sponsored Clinical Trials: Building a Site for Sharing Data and Models to evolve better Disease Maps.
Public-Private Partnership of leading pharmaceutical companies, clinical trial groups and researchers.
Neutral Conveners: Sage Bionetworks and Genetic Alliance [nonprofits].
Initiative to share existing trial data (molecular and clinical) from non-proprietary comparator and placebo arms to create powerful new tool for drug development.
Started Sept 2010
Shared clinical/genomic data sharing and analysis will maximize clinical impact and enable discovery
• Graphic of curated to qced to models
Arch2POCM
Restructuring the PrecompeQQve Space for Drug Discovery
How to potenQally De-‐Risk High-‐Risk TherapeuQc Areas
Arch2POCM: scale and scope
• Proposed Goal: Initiate 2 programs. One for Oncology/Epigenetics/Immunology. One for Neuroscience/Schizophrenia/Autism. Both programs will have 8 drug discovery projects (targets) - ramped up over a period of 2 years
– It is envisioned that Arch2POCM’s funding partners will select targets that are judged as slightly too risky to be pursued at the top of pharma’s portfolio, but that have significant scientific potential that could benefit from Arch2POCM’s crowdsourcing effort
• These will be executed over a period of 5 years making a total of 16 drug discovery projects
– Projected pipeline attrition by Year 5 (assuming 12 targets loaded in early discovery)
• 30% will enter Phase 1 • 20% will deliver Ph 2 POCM data 52
The FederaQon
2008 2009 2010 2011
How can we accelerate the pace of scientific discovery?
Ways to move beyond “traditional” collaborations?
Intra-lab vs Inter-lab Communication
Colrain/ Industrial PPPs Academic Unions
(Nolan and Haussler)
sage federation: model of biological age
Faster Aging
Slower Aging
Clinical Association - Gender - BMI - Disease Genotype Association Gene Pathway Expression Pr
edicted Age (liver expression)
Chronological Age (years)
Age Differential
Reproducible science==shareable science
Sweave: combines programmatic analysis with narrative
Sweave.Friedrich Leisch. Sweave: Dynamic generation of statistical reports using literate data analysis. In Wolfgang Härdle and Bernd Rönz,editors, Compstat 2002 –
Proceedings in Computational Statistics,pages 575-580. Physica Verlag, Heidelberg, 2002. ISBN 3-7908-1517-9
Dynamic generation of statistical reports using literate data analysis
Federated Aging Project : Combining analysis + narraQve
=Sweave Vignette Sage Lab
Califano Lab Ideker Lab
Shared Data Repository
JIRA: Source code repository & wiki
R code + narrative
PDF(plots + text + code snippets)
Data objects
HTML
Submitted Paper
TP53 mut
CDKN2A copy
MDM2 expr
HGF expr
CML linage EGFR mut
EGFR mut
EGFR mut
CML lineage
ERBB2 expr
BRAF mut
BRAF mut
NRAS mut
BRAF mut
NRAS mut
KRAS mut
BRAF mut
NRAS mut
KRAS mut
#1 BRAF mut
#2 NRAS mut #1 BRAF mut
#3 KRAS mut #2 NRAS mut #1 BRAF mut
#3 KRAS mut #2 NRAS mut #1 BRAF mut
#1 EGFR mut
#1 ERBB2 expr
#1 EGFR mut
#2 CML lineage #1 EGFR mut
#1 CML lineage
#1 HGF expr
#2 TP53 mut #3 CDKN2A copy #1 MDM2 expr
Can the approach make new discoveries?
For 11/12 compounds, the #1 predictive feature in an unbiased analysis corresponds to the known stratifier of sensitivity
59
Vaske, et al.
Presentation outline
Currently mRNA copy number somatic mutations (36
cancer-related genes) In progress targeted exon sequencing epigenetics microRNA lncRNA phospho-tyrosine kinase metabolites
Molecular characterization (1,000 cell lines)
Viability screens (500 cell lines, 24 compounds)
Small molecule screen
Cancer cell line encyclopedia
TCGA /ICGC Molecular characterization (50 tumor types)
genomics transcriptomics epigenetics
Clinical data Predic4ve model
1) Predic4ng drug response from cancer cell lines
2) Future approaches: network-‐based predictors and mul4-‐task learning
3) Standardized workflows for data management, versioning and method comparison
Transfer learning
Network / pathway prior informa4on
Vaske, et al.
1) Data management APIs to load standaridzed objects, e.g. R ExpressionSets (MaD Furia):
ccleFeatureData <-‐ getEnQty(ccleFeatureDataId) ccleResponseData <-‐ getEnQty(ccleResponseDataId)
tcgaFeatureData <-‐ getEnQty(tcgaFeatureDataId) tcgaResponseData <-‐ getEnQty(tcgaResponseDataId)
=!
Observed Data!=! +!
+!
Random Variation!Systematic Variation!
+!
Normalization: Remove the influence of adjustment variables on data...!
=! +!
2) Automated, standardized workflows for cura4on and QC of large-‐scale datasets (Brig Mecham).
A. TCGA: Automated cloud-‐based processing. B. GEO / Array Expression: NormalizaQon workflows, curaQon of phenotype using standard ontologies. C. AddiQonal studies with geneQc and phenotypic data in Sage repository (e.g. CCLE and Sanger cell line datasets)
custom model 1 custom model 2 custom model N
4) Sta4s4cal performance assessment across models.
custom model 1 custom model 2 custom model N
5) Output of candidate biomarkers and feature evalua4on (e.g. GSEA, pathway analysis)
6) Experimental follow-‐up on top predic4ons (TBD) E.g. for cell lines: medium throughput suppressor / enhancer screens of drug sensiQvity for knockdown / overexpression of predicted biomarkers.
3) Pluggable API to implement predic4ve modeling algorithms.
A) Support for all commonly used machine learning methods (for automated benchmarking against new methods)
B) Pluggable custom methods as R classes implemenQng customTrain() and customPredict() methods.
A) Can be arbitrarily complex (e.g. pathway and other priors)
B) Support for parallelizaQon in for each loops.
Portable Legal Consent
(AcQvaQng PaQents)
John Wilbanks
weconsent.us
Sage Congress Project April 20 2012
RealNames Parkinson’s Project RevisiQng Breast Cancer Prognosis
Fanconi’s Anemia
(Responders CompeQQons-‐ IBM-‐DREAM)
Networking Disease Model Building