The Role of Systems Modeling
in Drug Discovery
and Predictive Health
Eberhard O. Voit
I
B S
I
& Department of Biomedical Engineering
Atlanta, Georgia
Unither Nanomedicine & Telemedical Technology Conference
February 23-26, 2010
Quebec, Canada 1
Overview
The Grand Challenge of Systems Biology
The Drug Pipeline and its Challenges
Systems Modeling in Drug Discovery:
PBPK Models
Receptor Models
Pathway Models
Predictive Health
General Caveat: Most technical details will be skipped
2
“Grand Challenge” of Comp. Sys. Biol.
www.alternative-cancer.net/
dX 12
0,12 X 24 ,12, 24 X 300 ,12, 30 51,12 X 1251,12,12 X 28 ,12, 28 X 5151,12, 51 X 5251,12,52
f0 f f f 51 f f
dt
science.nationalgeographic.com
4,17 X 4 4,17, 4 X 22 ,17, 22 6,17 X 6 6,17,6 X 22,17, 22
f 41 f f 6 f
% model_ode
function dy = PP_ode(t,y)
…
dy(1) = y(1) * (b1 - b2 * y(1) - b3 * y(2));
dy(2) = y(2) * (b4 * y(1) + b5 - b6 * y(2));
end 3
www.ornl.gov/
“Grand Challenge” of Comp. Sys. Biol.
www.alternative-cancer.net/
dX 12
0,12 X 24 ,12, 24 X 300 ,12, 30 51,12 X 1251,12,12 X 28 ,12, 28 X 5151,12, 51 X 5251,12,52
f0 f f f 51 f f
dt
science.nationalgeographic.com
4,17 X 4 4,17, 4 X 22 ,17, 22 6,17 X 6 6,17,6 X 22,17, 22
f 41 f f 6 f
% model_ode
function dy = PP_ode(t,y)
…
dy(1) = y(1) * (b1 - b2 * y(1) - b3 * y(2));
dy(2) = y(2) * (b4 * y(1) + b5 - b6 * y(2));
end
4
www.ornl.gov/
Rewards for Solving the Grand Challenge
Testing incomparably faster and cheaper
Explain root causes of observations
Make reliable predictions
Understand design and operating principles
of biological systems
Create novel biological systems
Reduce animal experiments
Improve medicine
5
Drug Discovery Pipeline
Preclinical Clinical Postclinical
Discovery Development Development Development
Lead Clinical Clinical
Target ID Optimization Phase I Phase III Launch
Hit ID, Development Clinical FDA
Lead ID of Drug Phase II Approval
Candidate Process
Note: 10-20 years; ~ 1 Billion $ 6
Models in Drug Discovery:
Structural Biology, Bioinformatics
TI Hit Lead DC CP1 CP2 CP3 FDA L!!
NCE (New chemical entity) screening
QSAR
Binding prediction (molecular dynamics)
7
Models in Drug Discovery: PBPK
TI Hit Lead DC CP1 CP2 CP3 FDA L!!
Blood
k0B
Brain
kLB
kL0 L B IN
Lung
kBL
Fat kB0
Kidney
ADME: Absorption, Distribution,
Liver
Metabolism, and Excretion;
Extrapolation; Routes; Dosage
8
Drug Discovery: Receptor Binding
TI Hit Lead DC CP1 CP2 CP3 FDA L!!
? ?
Receptor Antibody Ligand
9
Receptor Binding
100
L0=100
C3
L0=10
L [%]
50
L0=0.01
L0=1
k+
3 k+ L
Inject 0
A k3– L 0 10
time [days]
20
kA–
k– k– kL– 100
2 1
L0=100
k2+ k1+
R C3 50
C2 – C1
kR+ kR L0=10
L0=0.01
L0=1
0
0 10 20
time [days]
10
Drug Discovery: Systems Analysis
TI Hit Lead DC CP1 CP2 CP3 FDA L!!
Signal S
Signal
MAPKKK Layer
X' = - k1 X E S + k2 XE + k7 XP
XE' = k1 X E S - (k2+k3) XE
XP' = k3 XE - k7 XP
MAPKKK MAPKKK-P
XP
PP3
MAPKK Layer
Y' = - k1 Y XP + k2 YE + k7 YP
YE' = k1 Y XP - (k2+k3) YE
YP' = k3 YE - k7 YP - k4 YP XP + k5 YPE + k8 YPP
MAPKK MAPKK-P MAPKK-PP YPE' = k4 YP XP - (k5+k6) YPE
YPP' = k6 YPE - k8 YPP
PP2 PP2
YPP
MAPK Layer
Z' = - k1 Z YPP + k2 ZE + k7 ZP
MAPK MAPK-P MAPK-PP ZE' = k1 Z YPP - (k2+k3) ZE
ZP' = k3 ZE - k7 ZP - k4 ZP YPP + k5 ZPE + k8 ZPP ZPP
PP1 PP1 ZPE' = k4 ZP YPP - (k5+k6) ZPE
ZPP' = k6 ZPE - k8 ZPP
Response Response
11
Drug Treatment
R5P
vprpps
vpyr
PRPP Ade
vade
(Purine Metabolism)
vgprt Pi
vden vpolyam
vhprt vaprt
SAM
Suppose too much UA
vmat
vtrans
vgmps
XMP
vimpd
IMP
vasuc
S-AMP
1. Explain:
Pi
e.g., PRPPS superactivity
GMP Ado
vgmpr vasli
GDP vampd AMP
ADP
GTP
or, HGPRT deficiency
ATP
vrnag vrnaa
RNA
Pi vgdrnr vgrna varna vadrnr
vgnuc
vgprt
dGMP
vdnag vdnaa dAdo 2. Intervene:
dAMP vada
reduce UA production
dGDP DNA dADP
dGTP
dATP
vgdna vadna
vdgnuc vdada
Pi vhprt
vinuc
HX
3. Side effects?
Gua
Guo Ino
vgua vhxd
dGuo Xa dIno
vx vxd
e.g.: UA Xa
UA vhx
vua
12
The Task of Personal Medicine
Status quo: Medicine is based on averages (either
from epidemiology or from animal experiments)
Task: Need to progress from average input-output
correlations to a deeper understanding of
disease processes in individuals
Challenges: 1. Get data
2. Analyze them appropriately
(i.e., per (sophisticated) modeling)
Hope: Analogy with engineering
We do not need to take apart every machine
we encounter, if we understand the principles
that make this type of machine functional.
13
Modeling Approaches
Biomarker associations (statistics, epidemiology)
Biomarker networks (biomarkers predictive of later
biomarkers; graph theory)
Parameter variations in disease models
time
Long-term development of health and disease
simplexes; personalized disease trajectories
14
Biomarker Modeling
One Biomarker:
(A to T) - SNP in HgbS Sickle Cell Anemia
Many Biomarkers:
Oncotype DX Test (21 genes) Remission of Breast Cancer
Hierarchical Networks of Biomarkers:
Disease
15
Dynamic Disease Modeling
Develop “physiological” model (of a disease prone system)
Set up equations
Identify parameter values for average, healthy individuals
“Personalize” models by altering parameters
(singly or in combination) and study
“disease” outcome
(targeted changes or MC simulations)
Model diseases and develop (personalized)
countermeasures
16
Molecular Biology Personalized
Epidemiology Biochemistry
Experimental Disease Models
Systems Biology
Physiology
Hypothesized
Risk-Factor~Disease Model Design
Associations “Averaged”
Model
Physiological
Mechanism Personalized
Perturbation Simulation
Simulation
Process
Parameters Numerical
Solution Sensitivity, Health-Disease
Robustness Classification
Clinical
Trials Personalized Personalized
Risk Profile Health Model
Suggested
Prevention
“Averaged” Personalized Personalized
Treatment Health Prediction Treatment
17
Computational Systems Biology
Voit & Brigham, Open Path. J., 2008
Biomarkers, Health and Disease Simplexes
One dimension: “normal range” (“U-box”)
normal biomarker
Two dimensions: combined normal ranges
biomarker 2
normal
normal biomarker 1
18
Voit, Math. Biosci. (2009)
Where does the Simplex Come from?
Biomarkers, Health and Disease Simplexes
Two dimensions: combined normal ranges + constraints
(Two extremes are not tolerable;
compensation between variables)
Result: linear bounds (reasonable approximation)
biomarker 2
normal
normal biomarker 1
19
Biomarkers, Health and Disease Simplexes
Many dimensions: polygon becomes a simplex
Note: Simplex can be computed from a model
20
Classification of Helath & Disease
Ideal Solution (in full “biomarker space”):
Clear separation between health and disease simplexes
z
“Health
Simplex”
y
“Disease
Simplex”
x 21
Classification of Helath & Disease
Would like to say: x : sick
(like PSA > 4)
z
y
x 22
Classification of Helath & Disease
In reality, there is no unique because disease status
also depends on other biomarkers, such as y and z.
z
y
“Healthy”
“Don’t know”
x “Diseased”
Consequence: Looking at one biomarker insufficient. 23
Health and Disease Trajectories
Health Premorbidity
Temporary Treatable or
Illness (fever, Self-healing
dehydration, …) Disease
24
Health and Disease Trajectories
Health Premorbidity
Temporary Treatable or
Illness (fever, Self-healing
dehydration, …) Disease
25
Summary
Modeling complements experimental systems biology
Quite a few technical challenges (e.g., parameter
estimation), but potential is clear
Systems modeling can play roles at several points of
the drug pipeline
Models are needed to assess personalized states,
diagnoses, treatments, and predictions
26
Acknowledgments
The 2008 Crew
Funding: NIH, NSF, DOE, Woodruff Foundation,
University System of Georgia, Georgia Research Alliance
Information: www.bst.bme.gatech.edu 27