Exploring Compound Combinations in High Throughput Settings: Going Beyond 1D Metrics
1. Exploring
Compound
Combina1ons
in
High
Throughput
Se9ngs
Going
Beyond
1D
Metrics
Rajarshi
Guha,
Lesley
Mathews,
John
Keller,
Paul
Shinn,
Dongbo
Liu,
Craig
Thomas,
Anton
Simeonov,
Marc
Ferrer
NIH-‐NCATS
January
2013,
San
Diego
2. Outline
Why
combine?
Physical
infrastructure
&
workflow
Summarizing
and
exploring
the
data
hKp://origin.arstechnica.com/news.media/pills-‐4.jpg
4. How
to
Test
Combina1ons
• Many
procedures
described
in
the
literature
– Fixed
dose
ra[o
(aka
ray)
– Ray
contour
– Checkerboard
– Gene[c
algorithm
C5
C5,D5
C4
C4,D4
C3
C3,D3
C2
C2,D2
C1,D5
C1,D4
C1,D3
C1,D2
C1,D1
D5 D4 D3 D2 D1
C1
0
5. Scaling
Response
Surface
Screening
5e+07
Combination type
• Response
surfaces
imply
a
DxD
matrix
for
each
combina[on
• All
pairs
screening
is
imprac[cal
for
more
than
tens
of
compounds
• Instead
we
consider
N
compounds
versus
a
fixed
size
library
All pairs
Fixed library
Number of combinations
4e+07
Dose matrix size
4
6
10
3e+07
2e+07
1e+07
0e+00
250
500
750
Number of compounds
1000
6. Mechanism
Interroga1on
PlateE
• Collec[on
of
~
2000
small
molecules
of
diverse
mechanism
of
ac[on.
• 745
approved
drugs
• 420
phase
I-‐III
inves[ga[onal
drugs
• 767
preclinical
molecules
• Diverse
and
redundant
MOAs
represented
belinostat
HDAC inhibitor
Phase II
AMG-47a
Lck inhibitor
Preclinical
GSK-1995010
FAS inhibitor
Preclinical
JZL-184
MAGL inhibitor
Preclinical
JNJ-38877605
HGFR inhibitor
Phase I
Eliprodil
NMDA antagonist
Phase III
7. Mechanism
Interroga1on
PlateE
Top
10
enriched
GeneGo
pathway
maps
Development EGFR signaling pathway
Some pathways of EMT in cancer cells
Development VEGF signaling via VEGFR2 - generic cascades
Apoptosis and survival Anti-apoptotic action of Gastrin
Cell adhesion Chemokines and adhesion
Cytoskeleton remodeling TGF, WNT and cytoskeletal remodeling
Regulation of lipid metabolism RXR-dependent regulation of lipid metabolism via PPAR, RAR and VDR
Transcription PPAR Pathway
Translation Non-genomic (rapid) action of Androgen Receptor
Development VEGF signaling and activation
0
5
-log10(pValue)
10
15
8. Combina1on
Screening
Workflow
Run
single
agent
dose
responses
6x6
matrices
for
poten1al
synergies
10x10
for
confirma1on
+
self-‐cross
Acoustic dispense, 15 min
for 1260 wells, 14 min for
1200 wells"
9. Where
Are
We
Now?
• 238
screens
in
total
– 30
screens
against
full
MIPE3
or
MIPE4
• 200
cell
lines
– Various
cancers
– Mainly
human
Number of assays
150
100
50
0
0
500
1000
1500
Number of combinations
• Combined
with
target
annota[ons
we
can
look
at
combina[on
behavior
as
a
func[on
of
various
factors
2000
10. Screening
Challenges
• A
key
challenge
is
automated
quality
control
• Plate
level
data
employs
standard
metrics
focusing
on
control
performance
• Combina[on
level
is
more
challenging
– Single
agent
performance
is
one
approach
– MSR
across
all
combina[on
can
provide
a
high
level
view
– But
how
to
iden[fy
bad
blocks?
11. QC
Examples
• Inves[ga[ng
an[-‐malarial
combina[ons
• 300
10x10
combina[ons
in
duplicate
• 15
compounds
included
more
than
ten
[mes
-1.5
log IC50 (uM)
-2.0
-2.5
-3.0
-3.5
-4.0
Artemether
Artesunate
Dihydro
artemisinin
Halofantrine Lumefantrine
12. QC
Examples
Compound
• Single
agents
with
very
high
MSR’s
could
be
used
to
flag
combina[ons
containing
them
• Doesn’t
help
for
compounds
with
only
one
or
two
replicates
• S[ll
requires
manual
inspec[on
Freq
40
30
20
10
0
5
10
MSR
15
20
15. Exploring
Combina1on
Metrics
• We
implement
a
variety
of
metrics
to
characterize
synergy/addi[vity/antagonism
• Lots
of
possible
ques[ons
– How
is
a
metric
distributed
in
a
given
assay?
– How
does
a
metric
vary
with
cell
line?
– Do
metrics
correlate?
– How
does
a
certain
combina[on
behave
across
cell
lines?
17. Repor1ng
Combina1on
Results
• These
web
pages
and
matrix
layouts
are
a
useful
first
step
• Does
not
scale
as
we
grow
MIPE
• S[ll
need
to
do
a
beKer
job
of
ranking
and
aggrega[ng
combina[on
responses
taking
into
account
– Response
matrix
– Compounds,
targets
and
pathways
18. When
are
Combina1ons
Similar?
• Differences
and
their
aggregates
such
as
RMSD
can
lead
to
degeneracy
• Instead
we’re
interested
in
the
shape
of
the
surface
• How
to
characterize
shape?
– Parametrized
fits
– Distribu[on
of
responses
0.06
0.010
0.04
0.005
0.02
0.00
0.000
0
25
50
75
100
0
0.15
0.10
0.05
0.00
0
50
100
D, p value
25
50
75
100
19. Similarity
via
the
KS
Test
• Quan[fy
distance
between
response
distribu[ons
via
KS
test
– If
p-‐value
>
0.05,
we
assume
distance
is
0
9
density
• But
ignores
the
spa1al
distribu[on
of
the
responses
on
the
concentra[on
grid
6
3
0
0.00
0.25
0.50
D
0.75
1.00
20. Similarity
via
the
Syrjala
Test
• Syrjala
test
used
to
compare
popula[on
distribu[ons
over
a
spa[al
grid
density
– Invariant
to
grid
orienta[on
– Provides
an
empirical
p-‐value
• Less
degenerate
than
just
considering
1D
distribu[ons
10.0
7.5
5.0
2.5
0.0
0.00
Syrjala,
S.E.,
“A
Sta[s[cal
Test
for
a
Difference
between
the
Spa[al
Distribu[ons
of
Two
Popula[ons”,
Ecology,
1996,
77(1),
75-‐80
0.25
D
0.50
0.75
21. Ibru1nib
Combina1ons
For
DLBCL
• Primary
focus
is
on
inves[ga[ng
combina[ons
with
Ibru[nib
for
treatment
of
DLBCL
– Btk
inhibitor
in
Phase
II
trials
– Experiments
run
in
the
TMD8
cell
line,
tes[ng
for
cell
viability
Viable
Cells
(% DMSO)
Ibrutinib
MK-2206
Ibrutinib* +
MK-2206
Ibrutinib* (nM)
MK-2206 (µM)
Mathews-‐Griner,
Guha,
Shinn
et
al.
PNAS,
2014,
in
press
24. 52
136
150
184
217
322
339
384
165
163
371
139
116
145
194
241
327
125
82
143
164
215
254
361
0.00
0.02
0.04
0.06
0.08
Cluster
C4
cellular carbohydrate biosynthetic process
• Focus
on
sugar
metabolism
• Ruboxistaurin,
cycloheximide,
2-‐
methoxyestradiol,
…
• PI3K/Akt/mTOR
signalling
pathways
regulation of polysaccharide biosynthetic process
cellular macromolecule localization
peptidyl-serine phosphorylation
regulation of generation of precursor metabolites and energy
cellular polysaccharide metabolic process
glucan metabolic process
glucan biosynthetic process
regulation of glycogen biosynthetic process
glycogen metabolic process
0
1
-log10(Pvalue)
2
3
25. Combina1ons
across
Cell
Lines
• Cellular
background
affects
responses
• Can
we
group
cell
lines
based
on
combina[on
response?
26. Working
in
Combina1on
Space
• Each
cell
line
is
represented
as
a
vector
of
response
matrices
L
L
• “Distance”
between
two
,
cell
lines
is
a
func[on
of
the
,
distance
between
component
response
matrices
,
,
D ( L1, L2 ) = F({d1, d2 ,…, dn })
,
• F
can
be
min,
max,
mean,
…
1
2
=
d1
=
d2
=
d3
=
d4
=
d5
29. Exploi1ng
Polypharmacology
DCC-2036
PD-166285
GDC-0941
PI-103
GDC-0980
Bardoxolone methyl
AT-7519
AT7519
SNS-032
NCGC00188382-01
Lestaurtinib
CNF-2024
ISOX
• PD-‐166285
is
a
SRC
&
FGFR
inhibitor
• Lestaurnib
has
ac[vity
against
FLT3
Vargatef
Belinostat
PF-477736
AZD-7762
Src
Lyn
Lck
Flt-3
PDGFRb
PDGFRa
FGFR-4
FGFR-3
Chk1 IC50 = 105 nM
FGFR-2
FGFR-1
VEGFR-3
VEGFR-2
VEGFR-1
0
200
Potency (nM)
Hilberg,
F.
et
al,
Cancer
Res.,
2008,
68,
4774-‐4782
400
600
30. Predic1ng
Synergies
• Related
to
response
surface
methodologies
• LiKle
work
on
predic[ng
drug
response
surfaces
– Peng
et
al,
PLoS
One,
2011
– Jin
et
al,
Bioinforma1cs,
2011
– Boik
&
Newman,
BMC
Pharmacology,
2008
– Lehar
et
al,
Mol
Syst
Bio,
2007
• But
synergy
is
not
always
objec[ve
and
doesn’t
really
correlate
with
structure
32. Predic1on
Strategy
• Don’t
directly
predict
synergy
• Use
single
agent
data
to
generate
a
model
surface
• Predict
combina[on
responses
• Characterize
synergy
of
predicted
response
with
respect
to
model
surface
• Reduced
to
a
mixture
predic[on
problem
• Need
to
incorporate
target
connec[vity
33. Conclusions
• Use
response
surfaces
as
first
class
descriptors
of
drug
combina[ons
– Surrogate
for
underlying
target
network
connec[vity
(?)
• Response
surface
similarity
based
on
distribu[ons
is
(fundamentally)
non-‐parametric
• Going
from
single
-‐
chemical
space
to
combina[on
space
opens
up
interes[ng
possibili[es
• Manual
inspec[on
is
s[ll
a
vital
step