never miss a discovery -...
TRANSCRIPT
1© 2018 Riffyn Inc
Never Miss a Discovery ®
PREP MEDIA
THAW CELLS INOCULATE ADD TREATMENT INCUBATE IMMUNO
ASSAY
MEASURE CELL COUNT
0.19540.2122
0.2397
0.1823 10.589
10.897
2.6349
10.732
2.57322.7589
11.344
2.6349
2© 2018 Riffyn Inc
“Data that goes into a database should obey what I call the CAP principle. It should be complete, it should be accurate, and it should be permanent, so you never have to do it again.
Otherwise, there’s no progress.”
— Sydney Brenner, Nobel Laureate, 2004
3© 2018 Riffyn Inc
* Error rate = R&D results that fail to reproduce, leading to wasted resources, failed tech transfers and missed discoveries.
Life Sciences
Chemicals/Materials
Energy
Food/AG
Mining and Metals
Pulp, Paper, Textiles
Water/Wastewater
Other
0 50 100 150 200 250 300
Sources:2013 R&D Magazine Global Funding ForecastNSF Science and Engineering Indicators 2012
CIA World Fact BookPrinz, et al., Believe it or not: how much can we rely on published data on potential drug targets?, Nat. Rev. Drug Disc., 2011
Begley & Ellis, Raise Standards for Preclinical Cancer Research, Nature, 483, 2012Ioannidis & Khoury, Improving Validation Practices in “Omics” Research, Science, 334, 2011
Halford, The Second Annual State of Translational Research, Sigma-Aldrich / AAAS, 2014Freedman, et al., The Economics of Reproducibility in Preclinical Research, PLOS Biology, 2015
R&D Spend
$420BR&D losses
$105B>25%error rate*
Annual R&D Spend ($B)
The problem in science today: >$100B of lost R&D productivity each year
4© 2018 Riffyn Inc
“It often takes us 2-3 months to assemble the data from a single development batch”
- senior scientist global pharmaceutical company
5© 2018 Riffyn Inc
Researchers spend 80% of their time cleaning and organizing data, instead of learning from it
20%learning
80%cleaning
6© 2018 Riffyn Inc
Wrangler
7© 2018 Riffyn Inc
Data Wrangler
8© 2018 Riffyn Inc
“Assessing the impact of batch variations on product quality is quite challenging.”
“We want integrated data for rapid correlative analysis across workflow &sample genealogy.”
“We are trying to escape Excel hell.”
“Our existing systems are too inflexible to support our changing processes.”
Statements of pain in R&D
Animal DMPK studies
Biologics Bioprocessing
Screening / Fermentation
Antibody Bio-panning
R&D data issues• quality• access• integration• interpretation• system flexibility
9© 2018 Riffyn Inc
But we, as a society, are failing to
to teach the principles,develop the tools, &
build the culture
of quality in R&D
Clean data starts with quality experiments
10© 2018 Riffyn Inc
Examples from the front lines of scienceThe bad, the ugly and the hopeful
11© 2018 Riffyn Inc
Examples selected from stages of the typical biotech development lifecycle
animal DMPK studies
cell line screening
bioprocess process
development
downstream process
development
analytical chemistry
12© 2018 Riffyn Inc
Example 1: Unreliable microtiter plate assay
0
1000000
2000000
3000000
4000000
5000000
6000000
96 w
ell a
ssay
| ou
tput
| fil
trate
| st
arch
| Fl
uore
scen
ce |
valu
e
Fron
tia |
GH3
0_8
Xyl |
A.n
FAEA
Fron
tia |
GH3
0_8
Xyl |
A.o
FAE
Fron
tia |
GH3
0_8
Xyl |
A.o
GUE
Fron
tia |
GH3
0_8
Xyl |
A.o.
BX
Fron
tia |
GH3
0_8
Xyl |
A.o.
LPM
OFr
ontia
| G
H30_
8 Xy
l | E.
n AX
EFr
ontia
| G
H30_
8 Xy
l | H.
i AXE
2Fr
ontia
| G
H30_
8 Xy
l | Nu
llFr
ontia
| G
H30_
8 Xy
l | T.
l CE1
6Fr
ontia
| G
H30_
8 Xy
l | T.
r AXE
1Fr
ontia
| G
H30_
8 Xy
l | T.
t AXE
Fron
tia |
GH5
_21
Xyl |
A.n
FAEA
Fron
tia |
GH5
_21
Xyl |
A.o
FAE
Fron
tia |
GH5
_21
Xyl |
A.o
GUE
Fron
tia |
GH5
_21
Xyl |
A.o.
BX
Fron
tia |
GH5
_21
Xyl |
A.o.
LPM
OFr
ontia
| G
H5_2
1 Xy
l | E.
n AX
EFr
ontia
| G
H5_2
1 Xy
l | H.
i AXE
2Fr
ontia
| G
H5_2
1 Xy
l | T.
l CE1
6Fr
ontia
| G
H5_2
1 Xy
l | T.
r AXE
1Fr
ontia
| G
H5_2
1 Xy
l | T.
t AXE
Fron
tia |
Null |
A.n
FAE
AFr
ontia
| Nu
ll | A
.o F
AEFr
ontia
| Nu
ll | A
.o G
UEFr
ontia
| Nu
ll | A
.o. B
XFr
ontia
| Nu
ll | A
.o. L
PMO
Fron
tia |
Null |
E.n
AXE
Fron
tia |
Null |
H.i
AXE2
Fron
tia |
Null |
T.l
CE16
Fron
tia |
Null |
T.r
AXE1
Fron
tia |
Null |
T.t
AXE
Fron
tia |
VD C
ellu
lase
| A.
n FA
EAFr
ontia
| VD
Cel
lula
se |
A.o
FAE
Fron
tia |
VD C
ellu
lase
| A.
o G
UEFr
ontia
| VD
Cel
lula
se |
A.o.
BX
Fron
tia |
VD C
ellu
lase
| A.
o. L
PMO
Fron
tia |
VD C
ellu
lase
| E.
n AX
EFr
ontia
| VD
Cel
lula
se |
H.i A
XE2
Fron
tia |
VD C
ellu
lase
| T.
l CE1
6Fr
ontia
| VD
Cel
lula
se |
T.r A
XE1
Fron
tia |
VD C
ellu
lase
| T.
t AXE
Null |
Nul
l | Fr
ontia
Ezyme Combination
Plate 1Plate 3 Plate 2Plate 4
Assa
y Va
lue
cell line screening
1
Column (top row) / Dose (bottom row)
Variability Chart
96 w
ell a
ssay
| o
| filtr
ate
| Abs
orba
nce
0
0.05
0.1
0.15
0.2
0.25
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 96 well enzyme prep | o | enzyme mixture | column0 3 6 12 24 36 60 96 well enzyme prep | i | enzyme 1 | mass ratio
2
Assa
y Va
lue
Dose of Control Reagent
Dose Response
0.020.040.060.080.1
0.120.140.160.180.2
0.220.240.26
96 w
ell a
ssay
| o
| filtr
ate
| Abs
orba
nce
(AU
)
0 5 10 15 20 25 30 35 40 45 50 55 6096 well enzyme prep | i | enzyme 1 | mass ratio (ug/
mg)
3
13© 2018 Riffyn Inc
Example 2:“Regression to the mean” – screening hits don’t hold up over time
17-RBOS-002117-RSC-0012 Signal Peptide Robustness Testing
17-RSC-0013 Strain Testing17-RSC-007 ExoproteasesELN-17-RBOS-0014 1-190
ELN-17-RBOS-0014 191-380ELN-17-RBOS-0016ELN-17-RBOS-0017
ELN-17-RBOS-0020_AELN-17-RBOS-0020_C
ELN-17-RSC-0008ELN-17-RSC-0009 4994
ELN-17-RSC-0010test
TEST
6
8
10
12
14
16
<0A-
> H
PLC
Ana
lysi
s | o
| Fe
rmen
tatio
n |
etha
nol |
con
cent
ratio
n
Bridgeport/AMP Redfield Energy Redfield Energy/LpH
Valero Valero/AMP
<0A-> Prepare corn mash | i | corn mash | resource name
Collected aggregate experiment data set from Riffyn. Identified common controls
Assessed media lot-to-lot effects on control strains
Model fit to control for lot-effects and temperature effects. Most of variation in product titer attributable to these 2 factors, not the candidate strains
Loteffect Temperatureeffect
1 2
3
CONCLUSION: Candidate strains thought to be hits were not hits (indistinguishable from parent).
4
Baddata?(Excludedfromfurtheranalysis)
9.510.010.511.011.512.012.513.013.514.014.5
<0A-
> H
PLC
Ana
lysi
s | o
| Fe
rmen
tatio
n | e
than
ol |
conc
entra
tion
Leve
rage
Res
idua
ls
11.0 11.5 12.0 12.5 13.0 13.5 14.0<0A-> Prepare corn mash | i | corn mash |
resource name Leverage, P<.0001
10.0
10.5
11.0
11.5
12.0
12.5
13.0
13.5
14.0
14.5<0
A->
HPL
C A
naly
sis
| o |
Ferm
enta
tion
| eth
anol
| co
ncen
tratio
n Le
vera
ge R
esid
uals
31.0 31.5 32.0 32.5 33.0 33.5 34.0 34.5 35.0 35.5 36.0<0A-> Fermentation | i | incubator |
temperature Leverage, P<.0001
StrainExperiment 1Experiment 2Experiment 3...
Experiment 14
Lot 1 Lot 2 Lot 3 Lot 4 Lot 5
Lot
ProductTiter
Temp
ProductTiter
Strain 1Strain 2Strain 3...
Reference Strain 1
.
.
.
Reference Strain 2....Strain N
# Samples
14© 2018 Riffyn Inc
Site 1 Site 2
MEAN
MEAN
Test
ing
resu
lt
Testing Day 1
Testing Day 3
Testing Day 2
Example 3:Pre-clinical testing results don’t match across independent sites
Same compound and protocol across 2 R&D sites
animal DMPK studies
15© 2018 Riffyn Inc
2 L 2 0 0 , 0 0 0 L
YIEL
D
BEFORE
PD MFG
XX% yield drop$X lost product
12 months transfer time
0.5 L
Example 4:Fermentation scale-up—troublesome tech transfers to full-scale manufacturing
bioprocess process
development
analytical chemistry
16© 2018 Riffyn Inc
Example 4:Fermentation process development—initial state
Time(h)
Yield
HIGH NOISE -> Unknowable process behavior
proc
ess
CV
TypicalPDBatchBEFOREQUALITY
%RecoveryofStand
ard
2.5%CV
98%accuracy
Quality improvement campaign
Original product assay
%RecoveryofStand
ard
0.3%CV
99.9%accuracy
Redeveloped product assay
17© 2018 Riffyn Inc
Example 4: Fermentation process development—after quality campaign
Time(h)
Yield
HIGH NOISE -> Unknowable process behavior
proc
ess
CV
TypicalPDBatchBEFOREQUALITY
Time(h)
Yield
LOW NOISE -> Predictive process model
proc
ess
CV
TypicalPDBatchAFTERQUALITY
18© 2018 Riffyn Inc
2 L 2 0 0 , 0 0 0 L
YIEL
D
BEFORE
PD MFG
XX% yield drop$X lost product
12 months transfer time
0.5 L
Exact performance matchacross scales
3-4 months transfer time
2 L 2 0 0 , 0 0 0 L
YIEL
D
AFTER
PD MFG
0.5 L
Example 4: Fermentation scale-up: before & after quality improvement and data integration
19© 2018 Riffyn Inc
Example 5:High-throughput strain screening artifacts and unreproducible hits
96 well plate data
Assa
y va
lue
Microtiter plate #
Root cause
20© 2018 Riffyn Inc
6X reduction in screening error doubles rate of strain improvement
1X R&D pace
2007 2008 2009 2010
2X R&D pace
6X error reduction
Year
Source: Gardner, TS (2013) Trends in Biotech. 31:3, 123-125.
BEFORE VARIANCE REDUCTION
AFTER VARIANCE REDUCTION
30% relative error5% relative error
Prod
uct y
ield
of a
str
ain
2X
21© 2018 Riffyn Inc
Learning from historyReduced decision-making error delivers faster progress
Improved data quality and integration revolutionizes the auto industry
Total Quality Implementation
Toyota
Nissan
Average for GM, Ford, and Chrysler
22© 2018 Riffyn Inc
Lessons of manufacturing quality
Define
Measure
Analyze / Improve
Control
DMAIC (6-sigma)(Mfg)
Design
Measure
Analyze / Improve
Share
Iterate
TRANSLATION TO R&D(R&D)
23© 2018 Riffyn Inc
But we, as a society, are failing to
to teach the principles,develop the tools, &
build the culture
of quality in R&D
Clean data starts with quality experiments
24© 2018 Riffyn Inc
We need to recognize our experiments as measurement systems
and teach our future scientists accordingly
25© 2018 Riffyn Inc
Never Miss a Discovery™
Riffyn’s mission is to help scientists deliver research we can trust and build upon efficiently.