This document discusses how biomedical discovery is being disrupted by big data. Large genomic, phenotype, and environmental datasets are needed to understand complex diseases that result from combinations of many rare variants. However, analyzing large biomedical data is costly and difficult given the standard model of local computing. The document proposes creating large "commons" of community data and computing as an instrument for big data discovery. Examples are given of the Cancer Genome Atlas project, which has petabytes of research data on thousands of cancer patients, and how tumors evolve over time. Overall, the document argues that new models of shared biomedical clouds and commons are needed to enable cost-effective analysis of big biomedical data.

1. Biomedical
Clusters,
Clouds
and
Commons
Robert
Grossman
University
of
Chicago
Open
Cloud
Consor=um
October
24,
2014
DePaul
CDM
Research
Colloquium

2. Part
1:
Biomedical
discovery
is
being
disrupted
by
big
data

3. Genomic
Data
• DNA
sequence
• RNA
expression
• etc.
Phenotype
Data
• Biology
• Disease
• etc.
Environmental
Data
• Environmental
exposures
• Microbial
environment
• etc.
data
driven
discovery

4. We
can
produce
lots
of
data,
but
why
do
we
need
so
much?

5. The
Dark
MaQer
of
Genomic
Associa=ons
with
Complex
Diseases:
Explaining
the
Unexplained
Heritability
from
Genome-­‐Wide
Associa=ons
Studies
(NHGRI
Workshop
held
in
Feb
2-­‐3,
2009)
Genome
Wide
Associa=on
Studies
Were
Disappoin=ng
Where
is
the
missing
inheritability?

6. What
is
the
Gene=c
Origin
of
Common
Diseases?
allele frequency HIGHLOW
effect
size
WEAK
STRONG
Rare alleles
causing Mendelian
diseases
Common variants
implicated in
common diseases
with GWAS
?
Current
hypothesis:
common
diseases
result
from
the
combina=on
of
many
rare
variants.
This
requires
lots
of
data.
Missing inheritability –
GWAS by and large failed to
find the genetic origin of
common diseases

7. One
Million
Genome
Challenge
• Sequencing
a
million
genomes
would
likely
change
the
way
we
understand
genomic
varia=on.
• The
genomic
data
for
a
pa=ent
is
about
1
TB
(including
samples
from
both
tumor
and
normal
=ssue).
• One
million
genomes
is
about
1000
PB
or
1
EB
• With
compression,
it
may
be
about
100
PB
• At
$1000/genome,
the
sequencing
would
cost
about
$1B
• Think
of
this
as
one
hundred
studies
with
10,000
pa=ents
each
over
three
years.

8. Four
Ques=ons
1. What
is
the
same
and
what
is
different
about
big
biomedical
data
vs
big
science
data
and
vs
big
commercial
data?
2. What
instrument
should
we
use
to
make
discoveries
over
big
biomedical
data?
3. Do
we
need
new
types
of
mathema=cal
and
sta=s=cal
models
for
big
biomedical
data?
4. How
do
we
organize
large
biomedical
datasets
and
the
community
to
maximize
the
discoveries
we
make
and
their
impact
on
health
care?

9. The
standard
model
of
biomedical
compu=ng
is
also
be
disrupted
by
big
data.
What
instrument
do
we
use
to
make
biomedical
discoveries?

10. Standard
Model
of
Biomedical
Compu=ng
Public
data
repositories
Private
local
storage
&
compute
Network
download
Local
data
($1K)
Community
socware
Socware
&
sweat
and
tears
($100K)

11. Instruments
are
dropping
in
cost,
devices
are
prolifera=ng,
and
pa=ents
are
dona=ng
data.

12. We
have
a
problem
…
Image:
A
large-­‐scale
sequencing
center
at
the
Broad
Ins=tute
of
MIT
and
Harvard.
Explosive
growth
of
data
New
types
of
data
It
takes
over
three
weeks
to
download
the
TCGA
data
at
10
Gbps
and
requires
$M’s
to
harmonize
Analyzing
the
data
is
more
expensive
than
producing
it
Not
enough
money

13. Source:
Interior
of
one
of
Google’s
Data
Center,
www.google.com/about/datacenters/
A
possible
solu=on:
create
large
“commons”
of
community
data
and
compute.
(Think
of
this
as
our
instrument
for
big
data
discovery)

14. New
Model
of
Biomedical
Compu=ng
Public
data
repositories
Private
storage
&
compute
at
medical
research
centers
Community
socware
Compute
&
storage
“The
Commons”

15. Solution: The Power of the Commons
Data
The Long Tail
Core Facilities/HS Centers
Clinical /Patient
The Why:
Data Sharing Plans
The
Commons
Government
The How:
Data
Discovery
Index
Sustainable
Storage
Quality
Scientific
Discovery
Usability
Security/
Privacy
Commons == Extramural NCBI == Research Object Sandbox == Collaborative Environment
The End Game:
KnowledgeNIH
Awardees
Private
Sector
Metrics/
Standards
Rest of
Academia
So2ware
Standards
Index
BD2K
Centers
Cloud, Research Objects,
Business ModelsSource:
Philip
Bourne,
NIH

16. Part
2:
Biomedical
Clouds
and
Commons

17. • The
Cancer
Genome
Atlas
(TCGA)
is
a
large
scale
NIH
funded
project
that
is
collec=ng
and
sequencing
disease
and
normal
=ssue
from
500
or
more
pa=ents
x
20
cancers.
• Currently
about
12,000
pa=ents
are
available.
• There
is
about
4
PB
of
research
data
today
and
growing.

18. TCGA
Analysis
of
Lung
Cancer
• 178
cases
of
SQCC
(lung
cancer)
• Matched
tumor
&
normal
• Mean
of
360
exonic
muta=ons,
323
CNV,
&
165
rearrangements
per
tumor
• Tumors
also
vary
spa=ally
and
temporally.
Source:
The
Cancer
Genome
Atlas
Research
Network,
Comprehensive
genomic
characteriza=on
of
squamous
cell
lung
cancers,
Nature,
2012,
doi:10.1038/nature11404.

19. Clonal
Evolu=on
of
Tumors
Tumors
evolve
temporally
and
spa=ally.
Source:
Mel
Greaves
&
Carlo
C.
Maley,
Clonal
evolu=on
in
cancer,
Nature,
Volume
241,
pages
306-­‐312,
2012.

20. TNBC
ER+
Source:
White
Lab,
University
of
Chicago.
Tumors
have
genomic
signatures
that
can
stra=fy
diseases
so
we
can
treat
each
stratum
differently.

21. Cyber
Pods
• New
data
centers
are
some=mes
divided
into
“pods,”
which
can
be
built
out
as
needed.
• A
reasonable
scale
for
what
is
needed
for
a
commons
is
one
of
these
pods.
• Let’s
use
the
term
“cyber
pod”
for
a
por=on
of
a
data
center
whose
cyber
infrastructure
is
dedicated
to
a
par=cular
project.
Pod
A
Pod
B

22. The
Bionimbus
Protected
Data
Cloud

23. Analyzing
Data
From
The
Cancer
Genome
Atlas
(TCGA)
1. Apply
for
access
to
data
(using
dbGaP).
2. Hire
staff,
set
up
and
operate
secure
compliant
compu=ng
environment
to
mange
10
–
100+
TB
of
data.
3. Get
environment
approved
by
your
research
center.
4. Setup
analysis
pipelines.
5. Download
data
from
CG-­‐
Hub
(takes
days
to
weeks).
6. Begin
analysis.
Current
Prac2ce
With
Bionimbus
Protected
Data
Cloud
(PDC)
1. Apply
for
access
to
data
(using
dbGaP).
2. Use
your
exis=ng
NIH
grant
creden=als
to
login
to
the
PDC,
select
the
data
that
you
want
to
analyze,
and
the
pipelines
that
you
want
to
use.
3. Begin
analysis.

24. Open
Science
Data
Cloud
• 500
users
use
OSDC
resources
• 150
ac=ve
each
month
• Users
u=lize
between
1000
–
100,000+
core
hours
per
month
for
compu=ng
over
PB
scale
commons
of
data

25. Goals
of
the
OSDC
Commons
• Build
a
commons
to
store,
harmonize
and
analyze
exis=ng
biomedical
data
that
scales
to
5-­‐100+
PB
• The
commons
must
support
“permanent”
genomic
data,
clinical
data,
environmental
data,
social
media
data,
donated
data,
etc.
• The
commons
must
provide
an
interac=ve
system
for
researchers
and
eventually
pa=ents
to
upload
their
data.
• Researchers
should
be
able
to
“pay
for
compute.”
• Pa=ents
should
be
able
to
use
the
commons
to
inform
their
treatment.

26. The
Tragedy
of
the
Commons
Source:
GarreQ
Hardin,
The
Tragedy
of
the
Commons,
Science,
Volume
162,
Number
3859,
pages
1243-­‐1248,
13
December
1968.
Individuals
when
they
act
independently
following
their
self
interests
can
deplete
a
deplete
a
common
resource,
contrary
to
a
whole
group's
long-­‐term
best
interests.
GarreQ
Hardin

27. Cloud
1
Cloud
3
Data
Commons
1
Commons
provide
data
to
other
commons
and
to
clouds
Research
projects
producing
data
Research
scien=sts
at
medical
research
center
B
Research
scien=sts
at
medical
research
center
C
Research
scien=sts
at
medical
research
center
A
downloading
data
Community
develops
open
source
socware
stacks
for
commons
and
clouds
Cloud
2
Data
Commons
2

28. Cloud
1
Data
Commons
1
Data
Commons
2
Data
Peering
• Tier
1
Commons
exchange
data
for
the
research
community
at
no
charge.

29. OSDC
Commons
Architecture
Object
storage
(permanent)
Scalable
light
weight
workflow
Community
data
products
(data
harmoniza=on)
Data
submission
portal
Open
APIs
for
data
access
and
data
access
portal
Co-­‐located
“pay
for
compute”
“Permanent”
Digital
ID
Service
&
Metadata
Service
Devops
suppor=ng
virtualized
environments

30. Part
3:
Analyzing
Data
at
the
Scale
of
a
Cyber
Pod
(the
need
for
a
new
data
science)
Source:
Jon
Kleinberg,
Cornell
University,
www.cs.cornell.edu/home/kleinber/networks-­‐book/

31. Complex
models
over
small
data
that
are
highly
manual.
Simpler
models
over
large
data
that
are
highly
automated.
Small
data
Medium
data
GB
TB
PB
W
KW
MW
Big
data
Cyber
pods

32. Is
More
Different?
Do
New
Phenomena
Emerge
at
Scale
in
Biomedical
Data?
Source:
P.
W.
Anderson,
More
is
Different,
Science,
Volume
177,
Number
4047,
4
August
1972,
pages
393-­‐396.

33. Several
ac=ve
voxels
were
discovered
in
a
cluster
located
within
the
salmon’s
brain
cavity
(Figure
1,
see
above).
The
size
of
this
cluster
was
81
mm3
with
a
cluster-­‐level
significance
of
p
=
0.001.
Due
to
the
coarse
resolu=on
of
the
echo-­‐planar
image
acquisi=on
and
the
rela=vely
small
size
of
the
salmon
brain
further
discrimina=on
between
brain
regions
could
not
be
completed.
Out
of
a
search
volume
of
8064
voxels
a
total
of
16
voxels
were
significant.
Source:
Craig
M.
BenneQ,
Abigail
A.
Baird,
Michael
B.
Miller,
and
George
L.
Wolford,
Neural
correlates
of
interspecies
perspec=ve
taking
in
the
post-­‐mortem
Atlan=c
Salmon:
An
argument
for
mul=ple
comparisons
correc=on,
retrieved
from
hQp://prefrontal.org/files/posters/BenneQ-­‐Salmon-­‐2009.pdf.

34. 4.
Center
for
Data
Intensive
Science
at
the
University
of
Chicago

35. Center
for
Data
Intensive
Science
• New
center
whose
mission
is
data
science
and
the
data
intensive
compu=ng
that
is
required
to
support
it.
• Focus
on
applica=ons
in
biology,
medicine
and
health
care,
but
interested
in
all
of
data
science.
• CDIS
is
organizing
around
some
challenge
problems
and
building
an
open
data
and
open
source
ecosystem
to
support
big
data
science.
• Leveraging
“instruments”
such
as
the
Protected
Data
Cloud
and
the
Open
Science
Data
Cloud.
• Building
data
commons
for
the
research
community,
star=ng
with
a
data
commons
for
genomic
data.
• POV:
“more
is
different.”

36. Data
Science
Data
Intensive
Applica=ons
Data
center
scale
compu=ng
(“cyber
pods”)
Data
driven
discoveries
Data
driven
diagnosis
Data
driven
therapeu=cs
We
are
developing
a
socware
stack
that
scales
to
a
cyber
pod
and
cura=ng
“commons
of
data”
that
we
can
use
as
an
“instrument”
for
data
driven
discoveries

37. Biomedical
Commons
Clouds
(BCC)
• The
not-­‐for-­‐profit
Open
Cloud
Consor=um
is
developing
and
opera=ng
a
biomedical
commons
and
cloud
called
the
Biomedical
Commons
Cloud
or
BCC
• BCC
is
a
global
consor=um
that
includes
universi=es,
companies
and
medical
research
centers
• Please
let
me
know
if
you
are
interested

38. Source:
David
R.
Blair,
Christopher
S.
LyQle,
Jonathan
M.
Mortensen,
Charles
F.
Bearden,
Anders
Boeck
Jensen,
Hossein
Khiabanian,
Rachel
Melamed,
Raul
Rabadan,
Elmer
V.
Bernstam,
Søren
Brunak,
Lars
Juhl
Jensen,
Dan
Nicolae,
Nigam
H.
Shah,
Robert
L.
Grossman,
Nancy
J.
Cox,
Kevin
P.
White,
Andrey
Rzhetsky,
A
Non-­‐Degenerate
Code
of
Deleterious
Variants
in
Mendelian
Loci
Contributes
to
Complex
Disease
Risk,
Cell,
September,
2013
5.
Challenges

39. The
5P
Challenges
• Permanent
objects
• Cyber
Pods
that
scale
• Data
Peering
• Portable
data
• Support
Pay
for
compute

40. Challenge
1:
Permanent
Secure
Objects
• How
do
I
assign
Digital
IDs
and
key
metadata
to
“controlled
access”
data
objects
and
collec=ons
of
data
objects
to
support
distributed
computa=on
of
large
datasets
by
communi=es
of
researchers?
– Metadata
may
be
both
public
and
controlled
access
– Objects
must
be
secure
• Think
of
this
as
a
“dns
for
data.”
• The
test:
One
Commons
serving
the
cancer
community
can
transfer
1
PB
of
BAM
files
to
another
Commons
and
no
bioinforma=cians
need
to
change
their
code

41. Challenge
2:
Cyber
Pods
• How
can
I
add
a
rack
of
compu=ng/storage/
networking
equipment
to
a
cyber
pod
(that
has
a
manifest)
so
that
– Acer
aQaching
to
power
– Acer
aQaching
to
network
– No
other
manual
configura=on
is
required
– The
data
services
can
make
use
of
the
addi=onal
infrastructure
– The
compute
services
can
make
use
of
the
addi=onal
infrastructure
• In
other
words,
we
need
an
open
source
socware
stack
that
scales
to
cyber
pods.

42. Challenge
3:
Data
Peering
• How
can
a
cri=cal
mass
of
data
commons
support
data
peering
so
that
a
research
at
one
of
the
commons
can
transparently
access
data
managed
by
one
of
the
other
commons
– We
need
to
access
data
independent
of
where
it
is
stored
– “Tier
1
data
commons”
need
to
pass
community
managed
data
at
no
cost
– We
need
to
be
able
to
transport
large
data
efficiently
“end
to
end”
between
commons

43. Challenge
4:
Biomedical
Data
Portability
• We
need
an
“Indigo
BuQon”
to
safely
and
compliantly
move
our
biomedical
data
between
Commons.
• Similar
to
the
HHS
“Blue
BuQon.”

44. Challenge
5:
Pay
for
Compute
Challenges
–
Low
Cost
Data
Integra=on
• Today,
we
by
and
large
integrate
data
with
graduate
students
and
technical
staff
• How
can
two
datasets
from
two
different
commons
be
“joined”
at
“low
cost”
– Linked
data
– Controlled
Vocabularies
– Dataspaces
– Universal
Correla=on
Keys
– Sta=s=cal
methods

45. 2005
-­‐
2015
Bioinforma2cs
tools
&
their
integra2on.
Examples:
Galaxy,
GenomeSpace,
workflow
systems,
portals,
etc.
2010
-­‐
2020
Data
center
scale
science.
Interoperability
and
preserva=on/peering/
portability
of
large
biomedical
datasets.
Examples:
Bionimbus/OSDC,
CG
Hub,
Cancer
Collaboratory,
GenomeBridge,
etc.
2015
-­‐
2025
New
modeling
techniques.
The
discovery
of
new
&
emergent
behavior
at
scale.
Examples:
What
are
the
founda=ons?
Is
more
different?

46. Ques=ons?
46

47. Major
funding
and
support
for
the
Open
Science
Data
Cloud
(OSDC)
is
provided
by
the
Gordon
and
BeQy
Moore
Founda=on.
This
funding
is
used
to
support
the
OSDC-­‐Adler,
Sullivan
and
Root
facili=es.
Addi=onal
funding
for
the
OSDC
has
been
provided
by
the
following
sponsors:
• The
Bionimbus
Protected
Data
Cloud
is
supported
in
by
part
by
NIH/NCI
through
NIH/SAIC
Contract
13XS021
/
HHSN261200800001E.
• The
OCC-­‐Y
Hadoop
Cluster
(approximately
1000
cores
and
1
PB
of
storage)
was
donated
by
Yahoo!
in
2011.
• Cisco
provides
the
OSDC
access
to
the
Cisco
C-­‐Wave,
which
connects
OSDC
data
centers
with
10
Gbps
wide
area
networks.
• The
OSDC
is
supported
by
a
5-­‐year
(2010-­‐2016)
PIRE
award
(OISE
–
1129076)
to
train
scien=sts
to
use
the
OSDC
and
to
further
develop
the
underlying
technology.
• OSDC
technology
for
high
performance
data
transport
is
support
in
part
by
NSF
Award
1127316.
• The
StarLight
Facility
in
Chicago
enables
the
OSDC
to
connect
to
over
30
high
performance
research
networks
around
the
world
at
10
Gbps
or
higher.
• Any
opinions,
findings,
and
conclusions
or
recommenda=ons
expressed
in
this
material
are
those
of
the
author(s)
and
do
not
necessarily
reflect
the
views
of
the
Na=onal
Science
Founda=on,
NIH
or
other
funders
of
this
research.
The
OSDC
is
managed
by
the
Open
Cloud
Consor=um,
a
501(c)(3)
not-­‐for-­‐profit
corpora=on.
If
you
are
interested
in
providing
funding
or
dona=ng
equipment
or
services,
please
contact
us
at
info@opensciencedatacloud.org.