A Multi-formalism solver (M2SL) and XML model markup (DAEML):
Application to the Guyton models
S. Randall ThomasIR4M UMR8081 CNRS - Univ. Paris Sud 11
Orsay & Villejuif, FRANCEand
A. I. HernándezLTSI - INSERM U642
Université de Rennes 1. Rennes, France
OpenModelica WorkshopFebruary 6, 2012
Linköping1
1Monday, February 6, 12
Overview
• The Guyton CVS (CardioVascular System) model: foundation for a collaborative “core model” for the VPH (Virtual Physiological Human)• Quick description of its underpinnings• Our modularization, extensions, and • Global sensitivity analysis and Virtual Population
• A brief introduction to M2SL: Multiformalism Multilevels Simulation Library • Objectives• Structure• User interface
22Monday, February 6, 12
SAPHIR & BIMBO collaboratorsIR4M UMR8081
(CNRS Orsay & Villejuif)Randy Thomas
Rob Moss Thibault Grosse
Stana AgnesPierre MazièreJérôme Bazin
Boubacar BenzianSylvain Demey
MC ENSIIE(Orsay)
Brigitte GrauAnne-Laure LigozatAnne-Lise Minard
IBISC (Univ. Evry)Fariza Tahi
Farida ZehraouiNadia Abchiche
Tarek Melliti
LTSI INSERM U.642 Rennes
Alfredo HernandezVirginie LeRolle
David OjedaGuy Carrault
Mireille Gareau
INSERM U927Poitiers
Patrick HannaertFrançois Guillaud
Collaborators fromBIMBO project
(Lyon)François Gueyffier
Ivanny MarchandAlexandra Laugerotte
Thierry Dumont
TIMC/PRETA UMR 5525 CNRS GrenoblePierre Baconnier
Julie Fontecave-JallonPascale Calabrese
Enas Abdulhay
Univ. Paris VIJean-Pierre Françoise
NASARon White
33Monday, February 6, 12
SAPHIR & BIMBO collaboratorsIR4M UMR8081
(CNRS Orsay & Villejuif)Randy Thomas
Rob Moss Thibault Grosse
Stana AgnesPierre MazièreJérôme Bazin
Boubacar BenzianSylvain Demey
MC ENSIIE(Orsay)
Brigitte GrauAnne-Laure LigozatAnne-Lise Minard
IBISC (Univ. Evry)Fariza Tahi
Farida ZehraouiNadia Abchiche
Tarek Melliti
LTSI INSERM U.642 Rennes
Alfredo HernandezVirginie LeRolle
David OjedaGuy Carrault
Mireille Gareau
INSERM U927Poitiers
Patrick HannaertFrançois Guillaud
Collaborators fromBIMBO project
(Lyon)François Gueyffier
Ivanny MarchandAlexandra Laugerotte
Thierry Dumont
TIMC/PRETA UMR 5525 CNRS GrenoblePierre Baconnier
Julie Fontecave-JallonPascale Calabrese
Enas Abdulhay
Univ. Paris VIJean-Pierre Françoise
NASARon White
3
Collaborations (HumMod, G92) with
the group of Jiri Kofranek,
Charles University, Prague
3Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Parameter'DB:'QKDB/
QSDB'
Interac2ve''
Model'repository:''
Virtual'Kidney/'KidneyGrid'
Portal'
Detailed(replacement(modules(
Robust((mul34formalism(solver:(M2SL(
DAEML'
(&'CellML)'
XSLT'
converters'
"Guyton"'
ontology'
Ontology'tools'
AutoKInforma2on'extrac2on'
from'PDF'files'
OMIE:'Ontology'Mapping'within'an'Interac2ve'
and'Extensible'environment'
44Monday, February 6, 12
Core-model environment: Guyton CVS models
kidney muscles
circulatory
dynamics
capillary
membrane
dynamics
thirst
ADH
control
angiotensin
control
aldosterone
control
electrolytes
& cell
water
tissue fluids,
pressures,
gel red cells,
viscosity
autonomic
control
pulmonary
dynamics
local blood
flow
control
oxygen
delivery
heart rate…
heart
hypertrophy
Guyton, Coleman, Granger (1972) Ann. Rev. Physiol.
55Monday, February 6, 12
Core-model environment: Guyton CVS models
kidney muscles
circulatory
dynamics
capillary
membrane
dynamics
thirst
ADH
control
angiotensin
control
aldosterone
control
electrolytes
& cell
water
tissue fluids,
pressures,
gel red cells,
viscosity
autonomic
control
pulmonary
dynamics
local blood
flow
control
oxygen
delivery
heart rate…
heart
hypertrophy
Guyton, Coleman, Granger (1972) Ann. Rev. Physiol.
+ Multi-organ interactions and regulatory loops
55Monday, February 6, 12
Core-model environment: Guyton CVS models
kidney muscles
circulatory
dynamics
capillary
membrane
dynamics
thirst
ADH
control
angiotensin
control
aldosterone
control
electrolytes
& cell
water
tissue fluids,
pressures,
gel red cells,
viscosity
autonomic
control
pulmonary
dynamics
local blood
flow
control
oxygen
delivery
heart rate…
heart
hypertrophy
Guyton, Coleman, Granger (1972) Ann. Rev. Physiol.
+ Multi-organ interactions and regulatory loops- Non pulsatile model, focused on long-term response
55Monday, February 6, 12
Core-model environment: Guyton CVS models
kidney muscles
circulatory
dynamics
capillary
membrane
dynamics
thirst
ADH
control
angiotensin
control
aldosterone
control
electrolytes
& cell
water
tissue fluids,
pressures,
gel red cells,
viscosity
autonomic
control
pulmonary
dynamics
local blood
flow
control
oxygen
delivery
heart rate…
heart
hypertrophy
Guyton, Coleman, Granger (1972) Ann. Rev. Physiol.
+ Multi-organ interactions and regulatory loops- Non pulsatile model, focused on long-term response
-> Integrate a pulsatile model into the Guyton model
55Monday, February 6, 12
Core-model environment: Guyton CVS models
kidney muscles
circulatory
dynamics
capillary
membrane
dynamics
thirst
ADH
control
angiotensin
control
aldosterone
control
electrolytes
& cell
water
tissue fluids,
pressures,
gel red cells,
viscosity
autonomic
control
pulmonary
dynamics
local blood
flow
control
oxygen
delivery
heart rate…
heart
hypertrophy
Guyton, Coleman, Granger (1972) Ann. Rev. Physiol.
+ Multi-organ interactions and regulatory loops- Non pulsatile model, focused on long-term response
-> Integrate a pulsatile model into the Guyton model-> Integrate RAAS model into the Guyton model
55Monday, February 6, 12
Blood pressure regulation: multi-organ integrationmany systems are involved, at many scales
from Guyton, A. C. (1980). Circulatory Physiology III. Arterial Pressure and Hypertension. Philadelphia, W.B. Saunders.
regulatory systems act over different time scales
66Monday, February 6, 12
Blood pressure regulation: multi-organ integrationmany systems are involved, at many scales
from Guyton, A. C. (1980). Circulatory Physiology III. Arterial Pressure and Hypertension. Philadelphia, W.B. Saunders.
regulatory systems act over different time scales
and over different pressure ranges
77Monday, February 6, 12
Guyton (G92): Comprehensive Sensitivity analysis...
•I/O maps of the 25 modules (all SAPHIR teams)• For each module: plots of all output variables as function of each input, over
a relevant physiological range of values
• Comprehensive sensitivity analysis (IBISC team)•Sensitivity of 297 system variables to each of 96 selected parameters at 5
min., 1h, 1day, and 4 weeks (steady-state) are calculated• This is done for normal steady state and also (twice) for >1000 x 96
randomized "individuals" (Morris. 1991. "Factorial Sampling Plans for Preliminary Computational Experiments." Technometrics, 33(2): 161-174)
• We have thus:• the mean ± SD of the effect of each parameter on each variable, • estimates of the interactions among the parameter effects (covariance
analysis provides details), and• a virtual population of env. 500 000 randomized individuals, and
88Monday, February 6, 12
G92 global sensitivity analysis"heatplots" of means of elementary effects
variables
para
met
ers
para
met
ers
variables
Mean values of the normalized effect, (% change of vj wrt its steady state value), of a small change of each parameter (one-at-a-time, 10% of allowed range) on all variables. Effects are shown at four times after the parameter change, as marked. The graph is truncated at ± 1%.
Clearly, the patterns change with time after the parameter perturbations.
99Monday, February 6, 12
In addition to the sensitivity analysis, per se:A Large Population of "Virtual (Guyton) Individuals"
Randomized parameters --> analogous to "genotype"This results in a variety of virtual "phenotypes"
Not surprisingly, a large proportion of the virtual population is "hypertensive"
The differences between parameter values of the normotensive vs. hypertensive subpopulations may be
interesting…
1010Monday, February 6, 12
Virtual "Guyton-population": Parameters most implicated in high BP in the virtual population
192 000 "virtual individuals" with randomized parameter values: 109,266 were Hypertensive (MAP above 106 mmHg)
Parameters whose means increased or decreased by at least 5% in the hypertensive subpopulation compared to normotensive subpopulation
Increased by >5% in hypertensives: AARK! basic afferent arteriolar resistanceANCSN! sensitivity controller of AngII effectCPR!critical plasma protein concentration for protein destructionKORGN! gain of positive feedback Korner conceptLPPR! rate of liver protein production).
Decreased by >5% in hypertensives: AUTOK! rate of development of very rapid autoregulationAUV!blood volume shifted from unstressed to stressedCPF!pulmonary capillary filtration coefficientEARK! basic efferent arteriolar resistanceGFLC! glomerular filtration coefficient)
from: Hernandez, A. I., V. Le Rolle, D. Ojeda, P. Baconnier, J. Fontecave-Jallon, F. Guillaud, T. Grosse, R. G. Moss, P. Hannaert and S. R. Thomas (2011). "Integration of detailed modules in a core model of body fluid homeostasis and blood pressure regulation." Prog Biophys Mol Biol 107(1): 169-182
1212Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level•therapeutic target
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level•therapeutic target
•the impact on blood volume, via NaCl reabsorption
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level•therapeutic target
•the impact on blood volume, via NaCl reabsorption•the resulting effect on blood pressure
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level•therapeutic target
•the impact on blood volume, via NaCl reabsorption•the resulting effect on blood pressure•hormonal and nervous system feedbacks
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level•therapeutic target
•the impact on blood volume, via NaCl reabsorption•the resulting effect on blood pressure•hormonal and nervous system feedbacks•(including possible effects on expression of the target gene!)
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level•therapeutic target
•the impact on blood volume, via NaCl reabsorption•the resulting effect on blood pressure•hormonal and nervous system feedbacks•(including possible effects on expression of the target gene!)
1313Monday, February 6, 12
Target scenario: Hypertension—Defects of Distal Tubule NaCl reabsorption.How to model the gene-to-organism relationship?
We must account for:•the change of function at the molecular/ cell membrane level•therapeutic target
•the impact on blood volume, via NaCl reabsorption•the resulting effect on blood pressure•hormonal and nervous system feedbacks•(including possible effects on expression of the target gene!)
BUT keep execution time manageable!
1313Monday, February 6, 12
M2SL: Multiformalism Multilevels Simulation Library® A. Hernandez(Rennes)
ODE/PDE
FEM
Multi-Agent Models
Cellular Automata
M2SL
LTSI%
1414Monday, February 6, 12
Core-model environment: Guyton CVS modelsHow should we implement this model? kidney muscles
circulatory dynamics
capillary membrane dynamics
thirst
ADH control
angiotensin control
aldosterone control
electrolytes & cell water
tissue fluids,
pressures, gel
red cells,
viscosity
autonomic control
pulmonary dynamics
local blood flow
control
oxygen delivery
heart rate…
heart Hyper- trophy
Kidney
Guyton72
Circulatory dynamics
Autonomic control
•••
Circulatory dynamics (d)
Heart Circulation
Electrical Activity
Mechanical Activity
Hydraulic Activity
Systemic circulation
Pulmonary circulation
Handling of different
time-scales
1515Monday, February 6, 12
Modeling & simulation method/tool: M2SL
www.ltsi.univ-rennes1.fr/m2sl
• M2SL: Multi-formalism modeling library, based on a co-simulation approach • Object-oriented (C++)• Hierarchical structures for Models and simulators
Coupled Model
Coupled Model
Atomic Model
Atomic Model
Atomic Model
Model hierarchy
Continuous formalism Discrete formalism
Coordinator
Coordinator
Simulator Simulator
Simulator
Root Coordinator
Simulator hierarchy
Temporal synchronization: • Different synchronization strategies
Inter-module coupling • Interface-based methods • S-H, Interpolation, …
A. Defontaine, A. I. Hernández, Acta Biotheoretica, vol. 52, pp. 273-90, 2004A. I. Hernández, et al Progress in Biophysics and Molecular Biology, vol. 107, pp. 169-182, 2011.
1616Monday, February 6, 12
M2SL: Multiformalism Multilevels Simulation Library® A. Hernandez(Rennes)
Mul$%formalism-modeling-by--co%simula$on!
• !Hierarchical!structure!!• !Object1oriented!!• !Distributed!approach!
Input/outputcoupling
Temporal Synchronization
Coupled model Coordinator
Coupled model
Atomic model Atomic model
Atomic model Coordinator
Simulator Simulator
Simulator
Con$nuous'Formalism' Doscrete'Formalism'
Model'hierarchy' Simulator'hierarchy'
time
O2 I1
M2 M1
I1 O2
Synchronisation & simulation at fixed timestep
Synchronisation at fixed timestep & adaptive simulation
Synchronisation & simulation both adaptive
1717Monday, February 6, 12
M2SL: Multiformalism Multilevels Simulation Library® A. Hernandez(Rennes)
Mul$%formalism-modeling-by--co%simula$on!
• !Hierarchical!structure!!• !Object1oriented!!• !Distributed!approach!
Input/outputcoupling
Temporal Synchronization
Coupled model Coordinator
Coupled model
Atomic model Atomic model
Atomic model Coordinator
Simulator Simulator
Simulator
Con$nuous'Formalism' Doscrete'Formalism'
Model'hierarchy' Simulator'hierarchy'
time
O2 I1
M2 M1
I1 O2
Synchronisation & simulation at fixed timestep
Synchronisation at fixed timestep & adaptive simulation
Synchronisation & simulation both adaptive
1717Monday, February 6, 12
Simulation example of the G72 model with M2SL
A. Hernández, et al. PTRS-A v.367,pp 4923-4940, 2009.
Benchmark – Adaptive with fixed couplingSimulation of 2 min. of intense exercise
DT initialized to 1e-4, coupling = 2.5e-4, max abs. Err = 5e-13Execution time 3.2 secs : ~ 3 times faster than a standard fixed-step
Model Outputs match reference benchmark ∆T of each sub-model
0 5 100.2
0.4
0.6
0.8
1
1.2vud
0 5 1026
28
30
32
34
36
38
40pvo
0 5 100
2
4
6
8
10pmo
0 5 1090
100
110
120
130
140
150pa
0 5 100
1
2
3
4
5aup
0 5 105
10
15
20
25qlo
0 5 100
5
10
15
20bfm
0 5 100
1000
2000
3000
4000mmo
0 5 100
0.2
0.4
0.6
0.8
1
1.2
Hemodynamics
time (min)
δt a
,1
0 5 100
0.2
0.4
0.6
0.8
1
1.2
AldoControl
time (min)
δt a
,2
0 5 100
0.2
0.4
0.6
0.8
1
1.2
AngioControl
time (min)
δt a
,3
0 5 100
0.2
0.4
0.6
0.8
1
1.2
MuscleBloodFlow
time (min)
δt a
,4
0 5 100
0.2
0.4
0.6
0.8
1
1.2
LocalBFControl
time (min)
δt a
,5
0 5 100
0.2
0.4
0.6
0.8
1
1.2
AntiDHormone
time (min)
δt a
,6
0 5 100
0.2
0.4
0.6
0.8
1
1.2
HeartViciousCycle
time (min)
δt a
,7
0 5 100
0.2
0.4
0.6
0.8
1
1.2
CapillaryMembrane
time (min)
δt a
,8
0 5 100
0.2
0.4
0.6
0.8
1
1.2
GelProtein
time (min)
δt a
,9
0 5 100
0.2
0.4
0.6
0.8
1
1.2
HeartHypertrophy
time (min)
δt a
,10
0 5 100
0.2
0.4
0.6
0.8
1
1.2
KidneySaltOut
time (min)
δt a
,11
0 5 100
0.2
0.4
0.6
0.8
1
1.2
PlasmaTissue
time (min)
δt a
,12
0 5 100
0.2
0.4
0.6
0.8
1
1.2
PulmonaryDynamics
time (min)
δt a
,13
0 5 100
0.2
0.4
0.6
0.8
1
1.2
Electrolytes
time (min)
δt a
,14
0 5 100
0.2
0.4
0.6
0.8
1
1.2
AutonomicControl
time (min)
δt a
,15
1818Monday, February 6, 12
M2SL www.ltsi.univ-rennes1.fr/m2sl
M2SL Models and simulators
Parallel (MPI) Evolutionary
algorithm Library
Simulated annealing
Kalman and Markov Library
Batch Processor
Matlab mex-interface
Parameter identification Sensitivity analysis
Analysis of intracerebral EEGs: Wendling F., Hernández A. Journal of Clinical Neurophysiology, 2005
Analysis of the ANS: Le Rolle V., Hernández A. Modelling and Simulation in Engineering 2008
Analysis of cardiac strain signals: Le Rolle V., Hernández A. Art. Int. Medicine. 2008
Integration of renal function into a CVS model R. Thomas, et al. PTRS-A, vol. 366, pp. 3175–3197 2008.
Multiresolution integration of pulsatile heart into a CVS model A. Hernández, et al. PTRS-A, vol. 367, pp. 4923-4940, 2009Analysis of long-term pacemaker data V. Le Rolle. IEEE TBME, vol. 58, pp. 2982-2986, 2011.
1919Monday, February 6, 12
M2SL www.ltsi.univ-rennes1.fr/m2sl
Generation of an XML (DAEML) file
XSLT
M2SL (C++) BM
Makefile
JNI calls
M2SL: Native library
HTML
Matlab
2020Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
2222Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
Detailed(replacement(modules(
2222Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
Detailed(replacement(modules(
Robust''mul*+formalism'solver:'M2SL'
DAEML&(&&CellML)&
XSLT&converters&
2222Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
Detailed(replacement(modules(
Robust''mul*+formalism'solver:'M2SL'
DAEML&(&&CellML)&
XSLT&converters&
Parameter DB: QKDB/QSDB
2222Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
Detailed(replacement(modules(
Robust''mul*+formalism'solver:'M2SL'
DAEML&(&&CellML)&
XSLT&converters&
Parameter DB: QKDB/QSDB
Ontology(tools(
2222Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
Detailed(replacement(modules(
Robust''mul*+formalism'solver:'M2SL'
DAEML&(&&CellML)&
XSLT&converters&
Parameter DB: QKDB/QSDB
Ontology(tools(
"Guyton"(ontology(
2222Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
Detailed(replacement(modules(
Robust''mul*+formalism'solver:'M2SL'
DAEML&(&&CellML)&
XSLT&converters&
Parameter DB: QKDB/QSDB
Ontology(tools(
"Guyton"(ontology(
Auto%Informa,on-extrac,on-from-PDF-files-
2222Monday, February 6, 12
SAPHIR: Towards a modular “core model” environment
Detailed(replacement(modules(
Detailed(replacement(modules(
Robust''mul*+formalism'solver:'M2SL'
DAEML&(&&CellML)&
XSLT&converters&
Parameter DB: QKDB/QSDB
Ontology(tools(
"Guyton"(ontology(
Auto%Informa,on-extrac,on-from-PDF-files-
Interac(ve**Model*repository:**
Virtual*Kidney/*KidneyGrid*Portal*
2222Monday, February 6, 12