Transcript presentation in ppt - Center for the Study of Complex Systems

Agent Based Models of the Acute
Inflammatory Response:
Update on Development and
Future Directions
Swarmfest 2004, Ann Arbor, MI
May 11, 2004
Gary An, MD
Department of Trauma
Cook County Hospital
Acute Inflammatory
Response (AIR)





Initial defense and repair mechanism
Specialized cellular/molecular pathways
Diffusely distributed/Tissue Nonspecific
Activation is non-specific to insult
Precedes Adaptive Immune response
(self/non-self distinction=>Antibodies)
Systemic Inflammatory
Response Syndrome/Multiple
Organ Failure (SIRS/MOF)




Disease of the ICU => “Unexplored State”
Pathologic state of Acute Immune
Response (AIR)
Physiologic manifestations result from
endogenous mediators
Hyperinflammation vs. Immunesuppression =>Temporal and Spatial
Challenge of SIRS/MOF
Gap between Pathophysiology
and Diagnosis
 Gap between Mechanisms and
Treatment
 Gap between Basic Science and
Clinical Implementation
 Nonlinear Behavior => Complexity

AIR as a Complex System
Components
 Rules

Locality
 Emergent
Properties
 Unexpected
Behavior

Cells
 Cellular
Programming
 Membranes/
Receptors
 Organ Physiology
 SIRS/MOF

Applications of ABM to
AIR/SIRS/MOF

Base Global Model
– Pathophysiology
– Therapeutic Interventions

Specific Disease
Processes/Pathogens/Mechanism
– Cutaneous and Inhalational Anthrax

Basic Science Experiment Simulation
– Epithelial Permeability Model
ABM of Global Systemic
Inflammation






Endothelial/Blood interface
Activation/Propagation of Inflammation
Endothelial Cells and White Blood Cells
Dynamics of Pathophysiology
Proto-Testing Platform for Systemic
Therapies
Very Abstract!
Current Model of Global
Inflammation
Cell types
Cell Receptors and
Functions
Mediators
Endothelial cells,
neutrophils, monocytes,
TH0, TH1, TH2, bacteria,
white bl ood ce ll ge nerative
cells
L-selectin, E/P-selectin,
CD-11/18, ICAM, TNFr, IL1r, adhesion, migration,
respiratory burst,
phagocytosis, apoptosis
Endotoxi n, PA F, TNF, IL-1,
IL-4, IL-8, IL-10, IL-12, IFNg, sTNFr, IL-1ra, GCSF
Validation Strategies




Agent Rules=>Transparency wrt code
Behavior of Individual wrt global
response to injury=>Individual
Dynamics
Behavior of Population wrt cytokine
patterns=>Population Dynamics
Behavior of Population wrt outcome to
intervention=>Population Response
Individual Response
Dynamics

Four possible dynamics:
– Successful healing
– “Phase II” or Immune-suppressed
SIRS/MOF
– “Phase I” or Hyper-inflammatory SIRS/MOF
– Overwhelming insult/infection

Function of degree of Initial Insult
Population Dynamics:
Cytokine Profiles





Patterns of cytokine levels for a
population at a specific IIN
7 days simulated time
IIN generates 50% mortality
N=100
Pattern Oriented/Qualitative (Very
Large Range-not shown)
Population Response:
Simulating Anti-inflammatory
Interventions
Any mediator represented as a variable
can be manipulated
 Modified based on published effects
 No other modifications of the ABM other
than simulated intervention
 Results all generated prospectively

List of In-Silico Experiments
3 day anti-TNF (Reinhart)
3 day rhIL-1ra (Opal)
7 day GCSF (Root)
Smaller Clinical Trials 1 dose anti-CD18 (Rhee)
Phase III Clinical
Trials
Animal Studies
3 day combination antiTNF and IL-1ra (Remick)
Hypothet ical Multimodal Regimes
anti-CD-18/anti-TNF/IL1ra
GCSF/anti-TNF/IL-1ra
ABM of Anthrax Infection


Modification of Base Global Model
Specific Characteristics of B. anthracis
– Both Cutaneous and Inhalational Forms
– Reproduce effects of Toxin-Component
(Lethal Factor, Edema Factor and
Protective Antigen) knockout species of B.
anthracis
Time of Death Distributions in All
Modes
Basic Science ABMs





Basic Science Paradigm = Linear
analysis
Examine Component Sub-Systems
Improve efficiency of Basic Science
experiments
Guide further investigation
Modular Components of System-wide
Model
ABM of Epithelial Cell
Permeability: Structure




Based on model of Delude
Epithelial cell culture => Grid of Epi Cell
Agents
Agent rules => Tight Junction (TJ)
Formation
TJ status determines permeability
ABM of Epithelial Cell
Permeability: Results




Increased Permeability to NO/Proinflammatory cytokine mix
Blocked with NO scavenger/iNOS
inhibitor
Matches Basic Science results
Potential Modular Model
Uses of ABM of the AIR

Formalize Mental Models
– Functional Repository of Basic Science
Information
– Modular
– Community-dependent

Drug Engineering
– Identify targets for manipulation
– Use to pre-test a planned treatment
regimes => Multi-Modal regimes
Uses of ABM of the AIR cont.

Clinical Therapeutics Design
– Patient Population Sub-stratification
– Generate Cytokine Profiles => “Finer
Grained”

Theoretical Tool
– Mathematical characterization of
system to guide future therapies
– “Cross Platform” Validation
Future Development

Multi-Tissue Model
– Directional Flow
– Coagulation
– Multiple Organ Failure/Support

Modular Model
– Basic Science Models
– Community/Web-based
– “Functional Data-bank”
Complex Systems
Rules drive Local interactions
between individual components
 Feedback loops =>non-linearity
 Interaction dynamics result in metastable structures=> Emergence
 Hierarchies of Emergent properties
 Non-intuitive, paradoxical behavior

Agent Based Modeling (ABM)
System of Components=>Agents
 Agent Rule systems=>Basic Science
 Populations of agents in virtual world
 Runs = agent actions/interactions=>
Locality
 Multiple runs=Random Number
Generators=> basic science experiments
 Stochastic and Deterministic

Why use ABM to model
AIR/SIRS/MOF?





Lots of information about potential
agents (cells and molecules)
Process is driven by local interactions
Dynamics may be too complex for topdown modeling
Multiple possible levels of model
validation
Integration of Models => Total System
Doing Science with ABM

In-Silico Experiments => Virtual
control and experimental
populations
– Apply standard statistical tools
– Use Pattern Oriented Analysis


Formalize mental model
building/testing hypotheses
Develop Theories
Population Runs



Random number generators are
active=> Heterogeneity
Multiple runs at a specific IIN generates
a study “population”
Generates a “mortality rate” for a
particular IIN (Mortality at >80% Total
Damage)=>“Control Population”
Results of In-Silico Experiments
in Sterile Mode (n=100)
Model Run
Mortality Chi Square
Base
86%
Antibiotics only
37%
Abs/anti-TNF
39%
Significant,
p=.01 vs. No Abs
NS
Abs/rh-IL-1ra
41%
NS
Abs/anti-CD18
42%
NS
Abs/rhIL-1ra/anti38%
TNF
Abs/anti-CD18/rhIL- 45%
1ra/anti-TNF
NS
NS
Results of In-Silico Experiments
in Infectious Mode (n=100)
Model Run Mortality
Base
100%
Antibiotics only 40%
Abs/anti-TNF
Abs/rhIL-1ra
Abs/GCSF
Abs/rhIL1ra/anti-TNF
Abs/GCSF/rhIL
-1ra/anti-TNF
Chi-square
42%
39%
36%
37%
Significant, p=.01
vs No Abs
NS
NS
NS
NS
38%
NS
End Oxy Deficit Distributions in All Modes
What ABM is not!
NOT a replacement of current
techniques of scientific investigation.
=> “Software vs. Hardware”
 NOT a clinical tool to provide a
prognosis or determine a treatment
course for an individual patient*

Translation, Synthesis and
ABMs

Requires data from Basic Science
– “What we look for and find out.”

Places it into Synthetic framework
– “How do the pieces fit together.”

Uses Multiple Hierarchies
– “Little pieces make big pieces.”

We Do This Already!
– Mental Models => Software Engineering
“Theories of SIRS/MOF”





“Dynamic Equilibrium”=>Response is
appropriate, degree is not
Concept of “anatomic containment” and
“physiologic containment”
Identify “Amplifiers” of response
Importance of all aspects of the response
=> “If a mediator does a lot of different
stuff don’t mess with it.”
Supplementation, not Blockade (WBCs
smarter than ICU MDs)
Summary of Key Points



The Acute Inflammatory Response is a
complex system that cannot be fully
characterized using existing
techniques.
Agent Based Modeling is well suited to
modeling the Inflammatory Response.
ABM would be an useful adjunct to
existing techniques of investigation.