Insights from Genomics and Computational Biology that inform precision medicine. Given at City of Hope Comprehensive Cancer Center

1. Insights from Genomics and
Computational Biology that inform
Precision Medicine
Warren Kibbe, PhD
warren.kibbe@nih.gov
@wakibbe
March 6th, 2017

2. 2
1. Background
2. Data in Biomedicine
3. Data Sharing
4. Data Commons
5. Genomics and
Computation
Thanks to many folks for slides, but especially Jerry Lee

3. 3
In 2016 there were an estimated
1,700,000 new cancer cases
and
600,000 cancer deaths
- American Cancer Society
Cancer remains the second most common cause of
death in the U.S.
- Centers for Disease Control and Prevention

4. 4
In 2016 there are an estimated
15,500,000
cancer survivors in the U. S.

5. 5
Understanding Cancer
 Precision medicine will lead to fundamental
understanding of the complex interplay between
genetics, epigenetics, nutrition, environment and clinical
presentation and direct effective, evidence-based
prevention and treatment.

6. 6
(10,000+ patient tumors and increasing)
Courtesy of P. Kuhn (USC)
2006-2015:
A Decade of Illuminating the
Underlying Causes of Primary
Untreated Tumors Omics
Characterization
Cancer is a grand challenge
Deep biological understanding
Advances in scientific methods
Advances in instrumentation
Advances in technology
Data and computation
Cancer Research and Care generate
detailed data that is critical to
create a learning health system for cancer
Requires:

7. 7
(10,000+ patient tumors and increasing)
Courtesy of P. Kuhn (USC)
2006-2015:
A Decade of Illuminating the
Underlying Causes of Primary
Untreated Tumors Omics
Characterization

8. The primary goal of TCGA is to identify disease subtypes and generate a
comprehensive catalog of all cancer genes and pathways
Pilot phase started in 2006 -- 3 tumor types 500 T/N pairs each
Extended in 2009 -- 20 x 500 T/N + 15 x (50-150) T/N
Freely available genomic data enable researchers anywhere
around the world to make and validate important discoveries

9. 9

10. 10

11. 11

12. Biological Scales

13. 13
Tumor, Cancer, and Metastasis:
(Length-scale and Time-scale Matter)
“…>90% of deaths are caused by disseminated disease or metastasis…”
Gupta et. al., Cell, 2006 and Siegel et. al. CA Cancer J Clin, Jan/Feb 2016
5 year Relative Survival Rates (2016 report of 2005-2011 data)

14. 14
1997 20152001 20051998 20142002 20061999 2003 20112000 2004 2007 20122008 20132009 2010
10/23/2001
(~5 yrs old)
4/21/1997
1/9/2007
(~10 yrs old)
iPod (10GB max)
WinAMP(mp3)
iPhone (EDGE, 16 GB max)
9/16/1999
(~3 yrs old)
802.11b WiFi
4/3/2010
(~13 yrs old)
iPad (EDGE, 64 GB max)
4/23/2005
(~8 yrs old)
9/26/2006
(~9 yrs old)
7/15/2006
2/7/2007
Google
Drive
4/24/2012
(~15 yrs old)7/11/2008
(~11 yrs old)
iPhone 3G
(16 GB max)
9/12/2012
(~15 yrs old)
iPhone5 (LTE, 128 GB max)
Google
Baseline
7/14/2014
(~17 yrs old)
3/9/2015
(~18 yrs old)
Digital technology is changing rapidly

15. 15
$640M
(FY74)
$5.21 B
(FY16)
Cancer therapy is changing

16. 1618
Application of Cancer Genomics is changing

17. 17

18. 18

19. 19
Biology and Medicine are now data
intensive enterprises
Scale is rapidly changing
Technology, data, computing and IT are
pervasive in the lab, the clinic, in the
home, and across the population

20. 20

21. 21

22. 22
“...an advantage of machine learning
is that it can be used even in cases
where it is infeasible or difficult to
write down explicit rules to solve a
problem...”
https://www.whitehouse.gov/sites/default/files/whitehouse_files/microsites/ostp/NSTC/preparing_for_the_future_of_ai.pdf

23. 23

24. 24

25. 25
" there is great potential for new insights to come
from the combined analysis of cancer proteomic
and genomic data, as proteomic data can now
reproducibly provide information about protein
levels and activities that are difficult or impossible
to infer from genomic data alone ”
Douglas R. Lowy, MD
Acting Director of the National Cancer Institute, National Institutes of Health
5/25/2016

26. 26
Genomics is the beginning of
precision medicine, not the end!

27. 27
Plumbing, architecture, genomics

28. Cancer Research Data Ecosystem – Cancer Moonshot BRP
Well characterized
research data sets Cancer cohorts Patient data
EHR, Lab Data, Imaging,
PROs, Smart Devices,
Decision Support
Learning from every
cancer patient
Active research
participation
Research information
donor
Clinical Research
Observational studies
Proteogenomics
Imaging data
Clinical trials
Discovery
Patient engaged
Research
Surveillance
Big Data
Implementation research
SEERGDC

29. Data Commons Framework

30. 30
Data Commons Structure
DICOM, AIM
Amazon
Google
IBM
Imaging
Validator
Q/A
Proteomic
Validator
Q/A
Clinical Phenotype
Validator
Q/A
MOD Phenotype
Validator
Q/A
Pathology
Radiology
Mass
Spectrometry
Array
Data
Commons
Security
Visualization
Authentication
& Authorization
Genomic
Validator
Q/A Germline Pipelines
DNA, RNA Pipelines
EMRs, Clinical
Trials
Azure
Data Contributors and Consumers
Researchers PatientsCliniciansInstitutions
NCI Thesaurus
caDSR
NLM UMLS
RxNorm
LOINC
SNOMED

31. 31
Cancer Data Sharing
& Data Commons
• Support open science
• Support data reusability
• Aligned with Cancer Moonshot
• Part of Precision Medicine
• Reduce Health Disparities
• Improve patient access to clinical
trials
• Work toward a learning National
Cancer Data Ecosystem
Reduce the risk, improve early detection, outcomes and survivorship in cancer

32. Cancer Data Sharing Efforts
Signature Efforts Data
BRCA Challenge
Somatic variant sharing
Isolated genetic variants
No raw sequencing data
Precision medicine questions
Somatic variant sharing
Panel gene resequencing
Clinical response
Clinical trial
Public-private partnerships
Comprehensive genomics
Detailed clinical
phenotype data
Clinical trial access
Clinical/genomic data
aggregation
EHR data
Clinical sequencing
Clinical oncology standards
EHR data
Clinical sequencing

33. 33
Changing the conversation around data sharing
 How do we find data, software, standards?
 How can we make data, annotations, software, metadata accessible?
 How do we reuse data standards?
 How do we make more data machine readable?
NIH Data Commons
NCI Genomic Data Commons
National Cancer Data Ecosystem
Data Commons co-locate data, storage and computing infrastructure, and
frequently used tools for analyzing and sharing data to create an
interoperable resource for the research community.
*Robert L. Grossman, Allison Heath, Mark Murphy, Maria Patterson, A Case for Data Commons Towards Data Science as a
Service, to appear. Source of image: Interior of one of Google’s Data Center, www.google.com/about/datacenters/.

34. 34
Basic Ingredients of a Cancer Research Data Ecosystem
• Open Science. Supporting Open Access, Open Data, Open Source, and
Data Liquidity for the cancer community
• Standardization through CDEs and Case Report Forms
• Interoperability by exposing existing knowledge through appropriate
integration of ontologies, vocabularies and taxonomies
• Consistency of curation and capture of evidence (tissue types, recurrence,
therapy – including genomic, epigenomic, etc context)
• Sustainable models for informatics infrastructure, services, data, metadata

35. 35
GDC as an example of a new
architecture for storing and sharing
cancer data

36. 36
The Cancer Genomic Data Commons
(GDC) is an existing effort to standardize
and simplify submission of genomic data
to NCI and follow the principles of FAIR
– Findable, Accessible, Attributable,
Interoperable, Reusable, and Provide
Recognition.
The GDC is part of the NIH Big Data to
Knowledge (BD2K) initiative and an
example of the NIH Commons
Genomic Data Commons
Microattribution, nanopublications, tracking the use of
data, annotation of data, use of algorithms, supports
the data /software /metadata life cycle to provide
credit and analyze impact of data, software, analytics,
algorithm, curation and knowledge sharing
Force11 white paper
https://www.force11.org/group/fairgroup/fairprinciples

37. 37
The NCI Genomic Data Commons
• Unify fragmentary repositories
• Support the receipt, quality control, integration, storage, and
redistribution of standardized genomic data sets derived from cancer
research studies
• Harmonization of raw sequence both from existing and new cancer research
programs
• Application of state-of-the-art methods of generating derived genomic data
• Provide the foundation for:
• Identification of high- and low-frequency cancer drivers
• Defining genomic determinants of response to therapy
• Clinical trial cohorts sharing targeted genetic lesions

38. NCI Genomic Data Commons
 The GDC went live on June 6, 2016 with approximately 4.1 PB of data.
 577,878 files about 14194 cases (patients), in 42 cancer types, across 29 primary
disease sites, 400 clinical data elements
 10 major data types, ranging from Raw Sequencing Data, Raw Microarray Data, to
Copy Number Variation, Simple Nucleotide Variation and Gene Expression.
 Data are derived from 17 different experimental strategies, with the major ones
being RNA-Seq, WXS, WGS, miRNA-Seq, Genotyping Array and Expression Array.
 Foundation Medicine announced the release of 18,000 genomic profiles to the GDC
at the Cancer Moonshot Summit, June 29th, 2016
 The Multiple Myeloma Research Foundation announced it would be releasing its
CoMMpass study of more than 1000 cases of Multiple Myeloma on Sept 29, 2016.

39. 39
Content in the Genomic Data Commons
 TCGA 11,353 cases
 TARGET 3,178 cases
Current
 Foundation Medicine 18,000 cases
 Cancer studies in dbGaP ~4,000 cases
 Multiple Myeloma RF ~1,000 cases
Coming soon
 NCI-MATCH ~3,000 cases
 Clinical Trial Sequencing Program ~3,000 cases
Planned (1-3 years)
 Cancer Driver Discovery Program ~5,000 cases
 Human Cancer Model Initiative ~1,000 cases
 APOLLO – NCI, VA and DoD ~8,000 cases
~58,000 cases

40. Exome-seq
Whole genome-seq
RNA-seq
Copy number
Genome
alignment
Genome
alignment
Genome
alignment
Data
segmentation
1° processing
Mutations
Mutations +
structural variants
Digital gene
expression
Copy number
calls
2° processing
Oncogene vs.
Tumor suppressor
Translocations
Relative RNA levels
Alternative splicing
Gene amplification/
deletion
3° processing
GDC Data Harmonization
Multiple data types and levels of processing

41. Mutect2
pipeline
GDC Data Harmonization
Open Source, Dockerized Pipelines

42. Recovery
rate
(% true
positives) A0F0
SomaticSniper 81.1% 76.5%
VarScan 93.9% 84.3%
MuSE 93.1% 87.3%
All Three 96.4% 91.2%
GDC variant calling
pipelines
Wash U
Baylor
Broad
GDC Data Harmonization
Multiple pipelines needed to recover all variants

43. Development of the NCI Genomic Data Commons (GDC)
To Foster the Molecular Diagnosis and Treatment of Cancer
GDC
Bob Grossman PI
Univ. of Chicago
Ontario Inst. Cancer Res.
Leidos
Institute of Medicine
Towards Precision Medicine
2011

46. Discovery of Cancer Drivers With 2% Prevalence
Lung adeno.
+ 2,900
Colorectal
+ 1,200
Ovarian
+ 500
Lawrence et al, Nature 2014
Power Calculation for Cancer Driver Discovery
Need to resequence >100,000 tumors to
identify all cancer drivers at >2% prevalence

47. The NCI Cancer Genomics
Cloud Pilots
Understanding how to meet the
research community’s need to
analyze large-scale cancer
genomic and clinical data

48. 48
NCI Cancer Genomics Cloud Pilots
Democratize access to
NCI-generated genomic
and related data, and to
create a cost-effective
way to provide scalable
computational capacity
to the cancer research
community.
Cloud Pilots provide:
• Access to large genomic data sets without need to download
• Access to popular pipelines and visualization tools
• Ability for researchers to bring their own tools and pipelines to the data
• Ability for researchers to bring their own data and analyze in combination with existing genomic
data
• Workspaces, for researchers to save and share their data and results of analyses

49. 49
• PI: Gad Getz
• Google Cloud
• Firehose in the cloud including Broad best practices workflows
•http://firecloud.org
Broad Institute
• PI: Ilya Shmulevich
• Google Cloud
• Leverage Google infrastructure; Novel query and visualization
•http://cgc.systemsbiology.net/
Institute for
Systems Biology
• PI: Deniz Kural
• Amazon Web Services
• Interactive data exploration; > 30 public pipelines
•http://www.cancergenomicscloud.org
Seven Bridges
Genomics
Three NCI Genomics Cloud Pilots
Selection
Design/Build
I
Design/Build
II
Evaluation Extension
Sept 2016Jan 2016April 2015Sept 2014
Jan 2014

50. Broad Institute Cloud Pilot
• Targeted at users performing
analyses at scale
• Modeled after their Firehose
analysis infrastructure
developed for the TCGA
program
• Users can upload their own data
and tools and/or run the Broad’s
best practice tools and pipelines
on pre-loaded data

51. Institute for Systems Biology Cloud Pilot
51
PI / Biologist
web access
Computational
Research Scientist
Python, R, SQL
Algorithm Developer
ssh, programmatic
access
ISB-CGC Web App Google Cloud Console
Google APIs
ISB-CGC APIs
Compute
Engine VMs
Cloud
Storage
BigQuery Genomics
Local
Storage
ISB-CGC
Hosted Data
Controlled-Access Data
Open-Access Data User Data
• Closely tied with Google Cloud Platform tools including BigQuery, App Engine, Cloud
Datalab, Google Genomics, and Compute Engine
• Aggregated TCGA data in BigQuery allows fast SQL-like queries across the entire dataset
• Web interface allows scientists to interactively compare and define cohorts

52. Seven Bridges Genomics Cloud Pilot
• Built upon the SBG commercial
cloud-based genomics platform
• Graphical query interface to
identify hosted data of interest
• Includes a native
implementation of the Common
Workflow Language
specification and graphical
interface for creating user-
defined workflows

53. Workspace –
isolated environment for collaborative analysis
Data + Methods → Results
sample data and
metadata (e.g.
BAMs, tissue type)
algorithms
(e.g. mutation
calling)
Wiring logic
(e.g. use the exome
capture BAM)
executions and results
(e.g. run mutation caller v41
on this exact bam and track
results)
Slide courtesy of Broad Institute

54. GDC Acknowledgements
NCI Center for Cancer Genomics Univ. of Chicago
Bob Grossman
Allison Heath
Mike Ford
Zhenyu Zhang
Ontario Institute for Cancer Research
Lou Staudt
Zhining Wang
Martin Ferguson
JC Zenklusen
Daniela Gerhard
Deb Steverson
Vincent Ferretti
'Francois Gerthoffert
JunJun Zhang
Leidos Biomedical Research
Mark Jensen
Sharon Gaheen
Himanso Sahni
NCI NCI CBIIT
Tony Kerlavage
Tanya Davidsen

55. CGC Pilot Team Principal Investigators
• Gad Getz, Ph.D - Broad Institute - http://firecloud.org
• Ilya Shmulevich, Ph.D - ISB - http://cgc.systemsbiology.net/
• Deniz Kural, Ph.D - Seven Bridges – http://www.cancergenomicscloud.org
NCI Project Officer & CORs
• Anthony Kerlavage, Ph.D –Project Officer
• Juli Klemm, Ph.D – COR, Broad Institute
• Tanja Davidsen, Ph.D – COR, Institute for Systems Biology
• Ishwar Chandramouliswaran, MS, MBA – COR, Seven Bridges Genomics
GDC Principal Investigator
• Robert Grossman, Ph.D - University of Chicago
• Allison Heath, Ph.D - University of Chicago
• Vincent Ferretti, Ph.D - Ontario Institute for Cancer Research
Cancer Genomics Project Teams
NCI Leadership Team
• Doug Lowy, M.D.
• Lou Staudt, M.D., Ph.D.
• Stephen Chanock, M.D.
• George Komatsoulis, Ph.D.
• Warren Kibbe, Ph.D.
Center for Cancer Genomics Partners
• JC Zenklusen, Ph.D.
• Daniela Gerhard, Ph.D.
• Zhining Wang, Ph.D.
• Liming Yang, Ph.D.
• Martin Ferguson, Ph.D.

56. 56
Integrated data sets, interoperable
resources, harmonized data are
necessary for and enable
biologically informed cancer
predictive models

57. 57
Questions?
Warren Kibbe, Ph.D.
Warren.kibbe@nih.gov
@wakibbe

58. www.cancer.gov www.cancer.gov/espanol

59. 59
NIH Genomic Data Sharing Policy
https://gds.nih.gov/
Went into effect January 25, 2015
NCI guidance:
http://www.cancer.gov/grants-training/grants-
management/nci-policies/genomic-data
Requires public sharing of genomic data sets