Agent-based models in translational systems biology

Overview

Agent-based models intranslational systems biologyGary An,1,4 Qi Mi,2,4 Joyeeta Dutta-Moscato,3,4 and YoramVodovotz3,4∗

Effective translational methodologies for knowledge representation are needed inorder to make strides against the constellation of diseases that affect the worldtoday. These diseases are defined by their mechanistic complexity, redundancy,and nonlinearity. Translational systems biology aims to harness the power ofcomputational simulation to streamline drug/device design, simulate clinical trials,and eventually to predict the effects of drugs on individuals. The ability of agent-based modeling to encompass multiple scales of biological process as well asspatial considerations, coupled with an intuitive modeling paradigm, suggests thatthis modeling framework is well suited for translational systems biology. Thisreview describes agent-based modeling and gives examples of its translationalapplications in the context of acute inflammation and wound healing . 2009 JohnWiley & Sons, Inc. WIREs Syst Biol Med 2009 1 159–171

THE ‘TRANSLATIONAL’ CHALLENGEOF BIOLOGY’S MULTIPLE SCALES

The sheer volume of biomedical research threatensto overwhelm the capacity of individuals to

process this information effectively. In particular,there are significant barriers to the effective integrationof discovered knowledge that arise from both thenature of biological systems and the structure ofthe biomedical research community. What is neededis the development of effective methodologies forbreaching these barriers to understanding, or moreprecisely, there is an acute need to translate knowledgeboth ‘vertically’ from the bench to the bedside,and be able to link ‘horizontally’ across multipleresearchers from various disciplines focused ondifferent diseases. We note that the traditional useof the term ‘translational research’ refers primarily tothe ‘vertical’ movement of knowledge from bench

∗Correspondence to: [email protected] of Surgery, Northwestern University, Chicago, IL606112Department of Sports Medicine and Nutrition, University ofPittsburgh, Pittsburgh, PA 152603Department of Surgery, University of Pittsburgh, Pittsburgh, PA152134Center for Inflammation and Regenerative Modeling, McGowanInstitute for Regenerative Medicine, University of Pittsburgh,Pittsburgh, PA 15219

DOI: 10.1002/wsbm.045

to bedside. However, given the nested, multiscaleorganization of biological systems and the barriersto understanding arising from that structure, therestriction of the term ‘translation’ to a narrow focusbelies the scope of the challenge facing the biomedicalresearch community. Therefore, we will use the term‘translation’ to refer to the transcending of thesebarriers wherever they may manifest. Furthermore,it should be noted that there are different types of‘scale’ in biomedical research. These include multiplelevels of organization (gene → protein/enzyme →cell → tissue → organ → organism),1 or ofabstraction.2 For instance, knowledge is generatedfrom research at multiple levels of organization,and the existence of these multiple levels presentssignificant challenges to the movement, application,and translation of mechanistic knowledge generatedat one organizational level to phenomena observedat a higher level. Furthermore, the organizationof the biomedical research community mirrors themultiscale structure of biological systems, wherespecialized research domains have evolved with focuson the processes at one particular level, but withlittle, if any, connection to adjacent processes andphenomena.1 This inherently fragmented structurehas led to a disparate and compartmentalizedresearch community and resultant disorganizationof biomedical knowledge. The consequences ofthis disconnect are seen primarily in difficulties indeveloping effective therapies for diseases resulting

Volume 1, September/October 2009 2009 John Wiley & Sons, Inc. 159

Overview www.wiley.com/wires/sysbio

from disorders of internal regulatory processes. Inthese settings, knowledge of internal mechanisticprocesses has not provided adequate insight intothe behavior of the system as a whole. Examples ofsuch diseases are cancer, autoimmune disorders andsepsis, all of which demonstrate complex, nonlinearbehavior. Moreover, attempts at developing therapiesfor these diseases have highlighted the epistemologicalbarriers to inferences of cause and effect thatresult from such complexity,3,4 and therefore requirethe development of translational methodologies toovercome these barriers.

AGENT-BASED MODELING: DYNAMICKNOWLEDGE REPRESENTATIONWhat is needed, then, is a means by which the mech-anistic information that is generated at one level ofbasic science research can be integrated with con-current parallel processes to produce recognizablephenomenological behaviors of the greater systemas a whole. This urgent need requires methods offormalizing the synthesis process of science, such thatdiscrete and partitioned knowledge can be representeddynamically to bring hypotheses to ‘life,’ which inthe clinical setting means the generation of improvedtherapies for the complex diseases that vex bothmodern, industrialized societies as well as developingnations. Mathematical modeling and computer simu-lation offer a method for achieving this translationalgoal.3,4 More specifically, computer modeling can beseen as a means of dynamic knowledge representationthat can form a basis for formal means of testing, eval-uating and comparing what is currently known withinthe research community.5 In this context, dynamiccomputational models can be considered a means of‘conceptual model verification,’ by which mental orconceptual models generated by researchers from theirunderstanding of the literature–—and used to guidetheir research–—can be instantiated computationallyand executed so that the behavioral consequencesof the researcher’s belief structure/hypothesis can beevaluated.5–7 This rapid cycle of prototyping and test-ing holds the promise of bringing to biomedicinethe precision and efficiency inherent to engineeringdisciplines.

Agent-based modeling is a rule-based, discrete-event and discrete-time computational modelingmethodology that employs computational objects thatfocuses on the rules and interactions among the indi-vidual components (‘agents’) of system.8–11 The goalof this type of simulation is the generation of pop-ulations of those system components and simulationof their interactions in a ‘virtual world’, to create an

in silico experimental model. It should be noted thatagent-based models (ABMs) are not inductive models,as are most mathematical and computational biomed-ical models, insofar that they are not based on patternsof data. Inductive models start with a set of data andmake inferences with respect to the mechanisms thatmight lead to that data. ABMs, on the other hand, startwith mechanisms or rules for behavior (albeit hypothe-ses in themselves) and seek to reconstruct through thecomputational instantiation of those mechanisms theobserved patterns of data. This is a critical point,because if the goal of a modeling endeavor is to findpatterns in an existing dataset, then agent-based mod-eling is likely not the appropriate method. However,if the goal is to test the veracity of a set of identi-fied/presumed mechanisms in a system, then agent-based modeling can be extremely useful. There areseveral characteristics of agent-based modeling that setit apart from other object-oriented, rule-based model-ing systems (such as Petri nets or Network models):

1. ABMs easily incorporate space. Agent-basedmodeling has its origins in two-dimensional cel-lular automata. As a result, many ABMs are‘grid-based.’ This legacy readily allows the spa-tial representation of the structural relationshipswithin a system as the two-dimensional gridhas been expanded into three dimensions5,12

and hexagonical space13,14 depending upon theneeds and goals of the developed model. Asa result, nonmathematicians can model fairlycomplex topologies with relative ease, leadingto more intuitive knowledge translation intoa simulation that can be manipulated, cali-brated to data, and verified. The spatial natureof ABMs also supports modeling agents with‘bounded knowledge,’ i.e., input constrained bylocality rules that determine its immediate envi-ronment. The emphasis on behavior driven bylocal interactions also matches closely with themechanisms of stimulus and response observedin biology. It should be noted, however, that notall ABM use physical space to define their virtualenvironments (for instance, ABMs of networkinteractions define the agent’s neighborhood viathe connection network rather than physicalproximity).

2. ABMs utilize parallelism. In an ABM, each agentclass has multiple manifestations as computa-tional objects forming a population of agentsthat interact in (a usually emulated) parallel pro-cessing environment. Differing local conditionslead to different behavioral trajectories of the

160 2009 John Wiley & Sons, Inc. Volume 1, September/October 2009

WIREs Systems Biology and Medicine Agent-based models in translational systems biology

individual agents, such that the heterogeneousbehavior of an individual agent within a pop-ulation of agents results in aggregated systemdynamics. These population dynamics are theobservable output of the ABM that representshigher-level system behavior. A classic exampleof this phenomenon is the behavior of flocks ofbirds, in which simulations utilizing relativelysimple interaction rules among birds can leadto sophisticated flocking patterns without anoverall controller.15

3. ABMs incorporate stochasticity. Many systems,particularly biological ones, include behaviorsthat appear to be random.16,17 ‘Appear to’ is animportant distinction, since what may appear tobe random at an observational level may be fullydeterministic from a mathematical standpoint.However, despite the fact that a particularsystem may follow deterministic rules, at theobservational level it is difficult or impossible todefine the rules or initial conditions within thesystem with only observation. ABMs addressthis issue via the generation of populations ofagents. Probabilities of a particular behavior aredetermined for the population as a whole, andresulting in the determination of a probabilityfunction for the behavior of a single agent, whichis in turn incorporated into the agent’s rules.As a population of agents is executed, eachagent follows a particular trajectory of behavioras its behavior rules’ probabilities collapse (i.e.are resolved) as the simulation progresses. Thisprocess enables the generation of a populationof behavioral outputs from a single ABM,producing system behavioral spaces consistentwith population-level biological observation.

4. ABMs have a modular structure. The behaviorof an ABM is largely defined by the rulesof its constituent parts, its agents. Newinformation can be added either through theintroduction of new agent-types or by themodification of existing agent rules withouthaving to reengineer the entire simulation. Forinstance, an ABM that is valid at one level ofabstraction, but with a relatively generic celltype, is expanded to include a finer degreeof detail with respect to subcategories of thegeneric cell type. To accomplish this newagent-classes and corresponding rules would beimplemented, but the remainder of the ABMwould remain essentially intact. Similarly, ifthere is a desire to include a new mediatorinto the model, this is done by creating a new

cellular-state or environmental variable, alongwith corresponding rules, without changingcore aspects of the original ABM. Finally,multiple ABMs can be aggregated, providingthat their points of contact and interactionare consistent across the incorporated ABMs(see the Functional Unit Representation Method(FURM) described by Hunt et al.14 and An’smultiscale gut–lung model5).

5. ABMs reproduce emergent properties. Becauseof the parallelism, intrinsic stochasticity, andlocally-constrained agent rules, a central hall-mark of ABM is that they generate systemicdynamics that could not have been reasonablyinferred from examination of the rules of theagents alone. This feature is termed emergentbehavior. To return to the example of the birdflock, superficial observation would seem to sug-gest the need for an overall leader to generateflock behavior, ostensibly requiring a meansof determining rules for flock-wide commandand control communication. This, however, isnot nature’s way; birds function on a seriesof locally-constrained interaction rules, and theflocking behavior emerges from the aggregateof these interactions.15 The capacity to gener-ate emergent behavior is a vital advantage ofusing ABM for conceptual model verification, asit is often the paradoxical, nonintuitive natureof emergent behavior that breaks a conceptualmodel.

6. ABMs can be constructed in the absenceof complete knowledge, keeping the rules assimple and verifiable as possible, even atthe expense of some detail. Meta-analysesof existing basic research provide structureand function to the ABM.18 This process ofmodeling also forces the question of how muchis really known. Hypothesis testing via ABMprovides qualitative verification of possibleoutcomes given incomplete knowledge. Sincemechanisms in biology are to some degreealways incompletely known, ABM maps wellto the means by which experimental biomedicalknowledge is currently expressed. Spatiallyinfluenced combinatorial freedom among agentsmeans that the emergent behavior has a rangeof stochasticity, similar to real biology. Theseoutcomes can be subjected to statistical analysisto assess qualitative trends based on the availableprior knowledge.10 It should be noted thatthere is a relationship between the fidelity ofmapping between the ABM and its biological



counterparts, and the strength of the correlationbetween the simulation result and the real-world behaviors. In general, a more detailedABM will lead to a stronger correlation tothe real-world and a greater confidence inthe ability of the ABM to describe observablephenomena. However, it should also be notedthat given the inherent incompleteness ofbiological knowledge, the relationship betweenthe ABM and the reference system’s behaviorwill likely never rise to the level of strictprediction.

Finally, a vital aspect of the use of agent-basedmodeling as an integrative modeling framework thatmoves toward the goal of communicable dynamicknowledge representation is the ease with whichbiomedical researchers can translate their conceptualmodels into executable form.5 While the era ofmultidisciplinarily trained researchers is dawning, itis still most likely that for the foreseeable futurethe majority of biomedical researchers will not beextremely facile in the use of computational toolsand methods. Agent-based modeling, by its object-oriented nature, has the advantage of mapping wellto the means by which biomedical knowledge iscurrently expressed, and is generally more intuitivefor nonmathematicians/computer scientists to usethan alternative modeling methods such as ordinarydifferential equations, partial differential equations,and their variants that allow for modeling stochasticprocesses (see below). As such, ABMs present alow barrier to entry for researchers who can bethus empowered to ‘bring to life’ their conceptualmodels, with subsequent comparison to traditionalin vitro and in vivo experimental results.19 Inaddition, since ABM are constitutive models, meaningthat they are constructed from the bottom-up toinstantiate mechanisms (as opposed to inductivemodels, in which mechanisms are inferred with thegoal of explaining data), the application of agent-based modeling arises from different questions thanequation-based inductive models. For instance, ABMsare not appropriate if the starting point is a mass ofraw data; rather, one must have already had some ideaof potential mechanisms that lead to the generationof the data. Therefore, one can envision an iterativeprocess by which inductive models are applied tolarge data sets, wet lab experiments are carriedout to evaluate and refine the mechanisms inferredfrom the inductive model, and the experimentallyconfirmed mechanisms are used as a basis of anABM which would close the discovery loop byrecapitulating the original data set. An excellent

example of this process can be found in a recent paperby Bailey et al., in which the authors developed amulticell ABM of human adipose-derived stromal celltrafficking during acute skeletal muscle ischemia.20

In this study, the authors showed that traffickingphenomena within cell populations emerged fromwell-vetted interactions among adhesion molecules,cytokines/chemokines, adhesion molecules, vascularhemodynamics, and the structure of the microvascularnetwork. Their ABM reproduced key aspects ofischemia and trafficking behavior. Importantly, theauthors’ simulations predicted the necessity of anunknown adhesion molecule to some of the observedphenomena, a prediction that was verified in vitro.Given the fact that intravenous delivery of humanadipose-derived stromal cells is a promising optionfor the treatment of ischemic injury, this study mayhave major translational implications.

TOOLS FOR AGENT-BASEDMODELING

Agent-based modeling has its origins in the fields ofecology, social science, and anthropology.8,10 Also,the necessary components of an agent-based modelingenvironment require addressing certain softwareissues not often dealt with in more traditionalobject-oriented programming tools. These issuesinclude means of emulating parallel processing, issuesrelated to concurrency, and the development ofschedulers to account for the multiple iterations thatconstitute an ABM run. Many ABMs are createdusing existing, general-purpose ABM developmentenvironments. These software packages are aimed atstriking a balance between representational capacity,computational efficiency, and user-friendliness.Among the most popular ABM toolkits are Swarm(http://www.swarm.org/index.php/Swarm main page),Mason (http://cs.gmu.edu/∼eclab/projects/mason/),RePast (http://repast.sourceforge.net/), NetLogo(http://ccl.northwestern.edu/netlogo/) and StarLogo(http://education.mit.edu/starlogo/). All these plat-forms represent some trade-off among the triad ofgoals mentioned above. For an excellent review andcomparison of agent-based modeling toolkits, seeRef. 21. An example of a RePast ABM can be seen inFigure 1, an example of a NetLogo-ABM is presentedin Figure 2, and Figure 3 demonstrates an ABMimplemented in Mason (see below).

Recently, a team at the University of Pittsburghhas developed a native ABM framework, SPARK(Simple Platform for Agent-based Representation of



(a) (b) (c)

FIGURE 1 | Description of the three zones of activity of the Basic Immune Simulator (BIS). (a) zone 1 is the parenchymal tissue zone. Thisrepresents a generic functional tissue (yellow circles represent parenchymal cell agents) that becomes infected with a virus (represented as the red,diffusing signal). If the average diameter of a cell to be approximated to be 0.01 mm, then zone 1 represents an area of approximately 1.0 mm2 oftissue. 1b zone 2 is the secondary lymphoid tissue zone. Secondary lymphoid tissue includes the lymph nodes and spleen. This is the site where thelymphoid cells, represented as B cell agents, T cell agents, and cytotoxic T lymphocyte agents, reside. This is also the site where agents representingantigen presenting cells (dendritic cell agents) interact with the lymphoid agents causing them to proliferate. (c) Zone 3 is the blood and lymphaticcirculation, and represents a transitional space between the site of initial infection (zone 1) and the lymphoid tissue (zone 2). As the agents in thesecondary lymphoid tissue proliferate (zone 2) they migrate into the lymph/blood (zone 3) and then travel back to the initial infection site (zone 1).This ABM was created using RePast software. (Reprinted with permission from Ref 22. Copyright 2007 http://www.biomedcentral.com).

A

B

FIGURE 2 | The multi-bilayer topology of the gut–lung ABM. Cubesand spheres are cell-level agents; these agents incorporatemolecular-level rule systems (scale-level 1). These rule systems result incellular behaviors (scale-level 2). Panel (a) is the pulmonary bilayer,with aqua cubes representing pulmonary epithelial cell agents, redcubes representing pulmonary endothelial cell agents, and below arespherical inflammatory cell agents. Panel (b) is the gut bilayer, with asimilar configuration, the only difference being that gut epithelial cellagents are pink. Each of these bilayers therefore represents anabstracted organ system (scale-level 3). Circulating inflammatory cellagents move between these two bilayers. Additionally, bilayer levelaggregated-variables representing inflammatory mesenteric lymph andblood-borne oxygen also move between the simulated gut and lung.This interconnection models the gut–lung axis, or multiorgan cross talk(scale-level 4). This ABM was created using NetLogo software. ABM,agent-based model. (Reprinted with permission from Ref 5. Copyright2008 http://www.biomedcentral.com).

Knowledge), a cross-platform freeware (download-able at www.pitt.edu/∼cirm/spark). SPARK is specif-ically designed as a biomedical ABM toolkit, and assuch focuses its functionality on the ABM rules sug-gested in Ref 1. A comparison of the above-mentionedABM software platforms is described in Table 1, andan example of a SPARK-based ABM is described inFigure 4 (see below).

USES AND APPROACHES TOAGENT-BASED MODELING

As noted above, agent-based modeling is only one ofan array of methods that have been used for dynamiccomputational simulations. These other methodsinclude equation-based modeling with ordinary orpartial differential equations, stochastic models suchas Markov chain Monte Carlo methods, and otherdiscrete-event models such as Petri Nets or NeuralNetworks. It is important to recognize that eachof these modeling techniques has its strengths andweaknesses, and that the modeling method chosenneeds to be tailored to the question being askedof the model.23 The scope of this article does notallow for the comprehensive comparison of all thesemethods; the focus is on the applicability of agent-based modeling in systems biology. Therefore, thefollowing section will list a series of ABMs that havebeen used in biomedical research, with particularattention to specific insights obtained through thechoice of utilizing agent-based modeling.



(a) (b) (c)EMT6/Ro Spheroids

Acceptable

Similarity

Hypothesized

Mappings

Mappings

Hypothesized

Mappings

Systemic Observations

and Data

Events: Local

Behaviors

Mechanisms: Cell-

Environment Interactions

Operating Principles

In Silico Analogue

Systemic Observations

and Data

Events: Local

Behaviors

Mechanisms: In Silico

Component Interactions

Cause Cause

Operating Principles

FIGURE 3 | Simulated spheroid tumor growth using axiomatic operating principles (rules). Panel (a) demonstrates the relationship between thesimulated multicellular tumor spheroids (SMS) and EMT6 spheroids (in vitro mouse mammary tumor cell cultures). An SMS is comprised of agentssimulating quasi-autonomous cell components interacting with surrounding agents and their environment using a set of axiomatic operatingprinciples (rules). There is a clear mapping between the SMS components and the EMT6 counterparts. Following execution, the interactingcomponents cause local and systemic behaviors. Measures of cell and system behaviors provide a set of attributes—the SMS phenotype. Theseattributes were calibrated to a target set of EMT6 attributes, suggesting a semiquantitative mapping between in silico and in vitro events. Panels(b) and (c) demonstrate the behavior of the SMS, with white circles = proliferating ‘cells’; light gray circles = quiescent ‘cells’; dark graycircles = ‘necrotic cells.’ The background gradient (from red to black) represent ‘nutrient’ levels relative to the maximum level in red. Panel(b) demonstrates growth in a high ‘nutrient’ environment, while panel (c) demonstrates growth in a low ‘nutrient’ environment. This ABM wascreated using Mason software. (Reprinted with permission from Ref 13. Copyright 2008. ABM, agent-based model.)

TABLE 1 List of features of Three ABM Frameworks

NetLogo Repast SPARK

Open source No Yes Yes

Multipleplatform

Yes Yes Yes

User target General Socialscience

Designed for biomedicalmodeling

Intracellular ModelingThe characterization of intracellular pathways andtheir behavior is the original focus of systems biology.As such, the systems biology community has a longhistory of work and achievement in the developmentof mathematical models of cellular signaling andmetabolic control. These models are generally basedon differential equations derived from experimentallydefined biochemical kinetics, and, more recently, havebeen executed using stochastic methods based onthe Gillespie Algorithm. The use of discrete-event,particle-based modeling exemplified by agent-basedmodeling has not been as well explored in this arena;however, there are certain insights that have beenobtained through the use of ABMs.

As mentioned above, the ability of ABMs toincorporate spatial relationships is a central bene-fit of their use, and this is particularly important inthe complex, compartmentalized environment of theintracellular milieu. As opposed to the well-mixed

systems found in a chemistry laboratory, enzymaticreactions within the intracellular environment areaffected by the relationships not only among thereactants but also with noninvolved molecules inter-posed between the reactants of interest. Moreover,the presence of some components in subcellular struc-tures, organelles, and compartments imposes furtherspatial constraints. Using an ABM of intracellularsignaling, Ridgway et al.24 demonstrated that the reac-tion dimension that determines biochemical kineticswithin a prokaryotic cytoplasm was reduced from theexpected three dimensions to nearly two. This findinghas significant consequences for the dynamic modelingof control loops in which subtle changes in feedbackdetermine the direction of a molecular switch. Inanother ABM of the control pathways affecting thetranscription factor nuclear factor kappa B (NF-κB),Pogson et al.25 demonstrated the need to account fornuclear translocation of the constitutive inhibitor ofNF-κB, I-kappa-B (IκB) based on the spatial distri-bution and activity of the reactions, particularly withrespect to the binding of IκB to actin. This mechanismhas been subsequently identified in their laboratory.26

Finally, it is recognized that cells are not merelybags of molecules dissolved in water, but rather thattheir cytoskeleton lends them a physical structure thatserves as a scaffold for enzymatic components of intra-cellular pathways. This spatial concept was imple-mented in an agent-based architecture called Spatially



1000(a)

(b) (c)

800

600

400

Vira

l cou

nt

200

0

0 1000 2000Number of steps after virus introduced

3000

FIGURE 4 | Chronic hepatitis C infection leads to hepatocellularcarcinoma. Panel (a) shows a model simulation of viral levels afterinfection with hepatitis C. As reflected in clinical presentation measuredby ALT levels, the typical course of infection shows an acute peak inviral load, followed by a low, chronic presence of the virus over a longperiod of time. Panel (b) shows the agents involved as inflammationprogresses. Green represent healthy hepatocytes, gray are infectedhepatocytes. As infected hepatocytes age and die, they become lighterin shade and turn white when dead, releasing virus (small black dots).The damage caused by infection and cell death provokes aninflammatory response mediated by macrophages (cyan circles), whichsecrete proinflammatory and anti-inflammatory cytokines. Thispromotes growth of the initial tumor cells (black mass in the center)over the course of years. Panel (c) shows hepatocellular carcinoma aftera few decades of chronic inflammation. Tumor-associated macrophagescluster within the tumor, which develops a hypoxic core with bothcancer and dead cells, and angiogenesis (red) as the mass becomesmalignant. This ABM was created using SPARK software. ALT, alaninetransaminase; ABM, agent-based model; SPARK, Simple Platform forAgent-based Representation of Knowledge.

Configured Stochastic Reaction Chambers and usedto demonstrate that even an abstract representation ofenzyme kinetics could, if sufficient component detailwas included within a spatial configuration, repro-duce canonical behavior at the cellular level, in thiscase the effect of preconditioning on the behavior ofthe toll-like receptor 4 (TLR-4) signaling pathway.27

Cell-level Tissue ModelingPerhaps the most obvious level of biologically-orientedagent-based modeling is that which uses cells asthe primary agent level. This type of component

representation provides an intuitive link between thelaboratory-derived basic mechanistic knowledge andthe structure of an ABM. It is therefore not surprisingthat the earliest examples of biomedically-relevantABMs are focused at this level.11,28 More detailswith respect to the implementation and integrationof agent-based modeling with wet lab experimentalprocedures can be seen in Refs 5,14,29,30.

One of the earliest insights via the use ofcell-level ABMs was the recognition that evenabstract agent rules could produce very recognizabledynamics, and in so doing provide vital insights intothe essential characterization of a disease process.To draw upon the example of modeling sepsis,an early ABM of systemic inflammation viewedthe inflammatory process as being mediated andpropagated by interactions at the endothelial bloodinterface.11 This ABM further demonstrated that thediffering trajectories of model-system behavior couldbe generated purely by altering the degree of initialperturbation, and that these four different trajectoriesmatched the four primary clinical scenarios associatedwith systemic inflammatory response. This ABMalso demonstrated that the mechanistic basis ofinflammation was the same whether the initiatinginsult was infectious, as in classical endotoxin sepsis,or tissue damage, as in massive trauma. These conceptshave gained general acceptance in recent years, but atthe time that this ABM was created these were noveland controversial hypotheses.

As with the intracellular ABMs, cell-level ABMsare well suited to modeling pathophysiologicalprocesses that have a significant spatial component.Examples of this utility can be seen in the modelingof granuloma formation,31 tumor growth,12,13,28,32

morphogenesis,30,33−35 angiogenesis,36 and woundhealing.37–39 A few highlights of this work aredescribed below.

In the first case, the propagation of Mycobac-terium tuberculosis was modeled to demonstrate gran-uloma formation resulting from the control processesinfluencing cell-to-cell interactions.31 In the case oftumor growth, the ABMs developed from the Deis-boeck laboratory demonstrated the importance of cell-to-cell interactions in generating the differential zonesof proliferation and senescence among tumor cells,thus also having implications with respect to the deliv-ery and efficacy of chemotherapeutic agents.12,28,40,41

A different approach to tumor growth by the Huntlab focused on using the ability of agent-based mod-eling to produce complex behaviors from abstractrules to identify key functions associated with tumorgrowth13,32 (more on this use of ABMs below). Sim-ilarly, the ability of agents to reconstruct complex



spatial configurations has been beneficial in the studyof morphogenesis. Cell-to-cell interactions alone havebeen experimentally identified as being sufficient tocreate morphological structures33 and ABMs havebeen able to reproduce these types of behavior insilico.30,33–35 The work of Peirce et al. on angio-genesis was able to use cellular response rules toreproduce patterns of neovascularization from vas-cular networks mapped from actual tissue.36 Finally,the simulations of wound healing were able to shedbasic insights on the spatial nature of the skin woundhealing process,37 to reproduce the impaired processesleading to the development of diabetic wounds andto posit potential mechanistic targets for therapeuticsdevelopment,38 and offer the potential for model-ing individual responses to injury and therapy in vocalchord trauma.39 In the impaired skin healing studies,38

a simulation of normal skin healing was used to deter-mine the phenotypic effects of diabetes-associatedderangements in single factors out of a complex cas-cade (namely underactivation of latent transforminggrowth factor-β1 or overproduction of tumor necrosisfactor-α). Simulating these single derangements wasassociated with a host of emergent features charac-teristic of diabetic ulcers. Moreover, this ABM wasused to test in silico the effects of both current ther-apies for diabetic ulcers (namely wound debridementand treatment with platelet-derived growth factor) aswell as novel interventions (e.g., inhibition of tumornecrosis factor-α or provision of transforming growthfactor-β1).38 In the last case, an ABM of vocal foldinflammation and healing was calibrated using dataon cytokine levels in laryngeal secretions of indi-vidual human volunteers subjected to experimentalphonotrauma (loud phonation). Patient-specific com-putational simulations were created based on baselinelevels of cytokines as well as levels at 1 and 4 hoursafter phonotrauma. These simulations were largelycapable of predicting the levels of cytokines at muchlater time points (24 hours), and were used as thebasis for simulated vocal fold therapy.39

Recently, an ABM of hepatitis C-inducedinflammation and hepatocellular carcinoma wascreated recently using SPARK (Dutta-Moscato et al.,manuscript in preparation). This model simulates theonset and progression of hepatocellular carcinomain a patch of liver, arising from only a fewcells (hypothetical cancer stem cells; Figure 4), inthe presence of a low, chronic inflammatory state[Figure 4(b)]. The tissue environment of pro- and anti-inflammatory responses present in chronic infections,such as hepatitis C virus [Figure 4(a)], causesprogressive tissue damage and accelerates tumorformation [Figure 4(b)]. Associated processes are also

modeled, including the formation of a hypoxic corein the center of the growing tumor, as well as tumorangiogenesis [Figure 4(c)].

These examples are far from comprehensive,but are intended to give a flavor of the types ofinflammation-related ABM that have been focused atthe tissue level. Below, we discuss the integration ofmodels at various scales.

Multiscale ApproachesAs noted above, biological systems exhibit a multiscaleorganization, and transcending these scales is aprimary goal of dynamic computational modeling.From a translational perspective, the payoff tolinking processes at multiple scales lies in bridgingstudies of therapeutic targets at the molecular/cellularlevel to the effects of modulation of these targetsat the tissue, organ, and whole-organism levels.While a two-scale accounting is intrinsic to mostmodeling methods, in which scale 1 = mechanismand scale 2 = mechanism consequence/output, agent-based modeling is particularly well suited to multiscalerepresentation, as this type of modeling automaticallytranscends these two scales and can account for athird scale of behavior, namely population dynamicsof parallel mechanism consequences/output. Indeed,there are multiple examples of multiscale modelingapproaches using an ABM framework.

One approach to representing multiple scales isto view sequential levels from an anatomic standpoint.The Basic Immune Simulator (BIS) developed byFolcik et al.22 is an example of this approach. TheBIS incorporates multiple tiers of tissue spaces whereimmunological processes occur. These tiers are labeled‘zones’ in the BIS, and consist of zone 1, whichrepresents generic parenchymal/peripheral tissue thatis subjected to an infectious insult, zone 2, whichrepresents regional lymphoid tissue in which the initialspecialized immunological response occurs, and zone3, a circulatory compartment that provides a conduitfor immune-activated cells from zone 2 to migrateback to the site of initial insult in zone 1 (Figure 1).This model reproduces the general dynamics of viralinfection, including a ‘hyperimmune’ response thatis analogous to acute hypersensitivity reaction. TheBIS is quite abstract, as it is intended to be a basicplatform that can be shared and modified as individualresearchers incorporate and test their own hypotheses.Interested readers are encouraged to examine theprimary text and download the application.

The structural/anatomic approach to multiscalemodeling is taken one step further with the intro-duction of a modular multiscale ABM architecture



focusing on acute inflammation.5 This architectureincorporates multiple structural and anatomic spaces(e.g., endothelial and epithelial surfaces, aggregatedcell-types into organ-specific tissues and finally organ-to-organ interconnections and cross talk). Further-more, this architecture also translates knowledgeacross domain specialties, such as capturing mech-anisms and behaviors derived from in vitro studies,leading to ex vivo tissue experiments and observa-tions, leading to patterns of organ-specific physiology,and finally to a clinically relevant, interconnected,multiorgan physiological and pathophysiological axis(Figure 2). Interested readers are encouraged to exam-ine the referenced text for additional details.

As mentioned above, differing biological pro-cesses may warrant differing modeling approaches,and this is especially true in terms of the differingcharacteristics of the multiple levels present in a mul-tiscale model. Utilizing the advances and resourcesof the traditional systems biology community, Athaleet al. have incorporated differential equation modelsdescribing the behavior of epidermal growth factorreceptor (EGFR) as the rule sets for a cell-level ABM,and used this model to examine the dynamics oftumor growth in various cancers.41,42 Similarly, thebehaviors of certain cell populations can be modeledmore efficiently using differential equations, leading tohybrid models. One such model, described by Wake-land et al.43 takes the previously described ABM ofsystemic inflammation and applies a systems dynam-ics model (ordinary differential equation) to moreefficiently represent the dynamics of the inflammatorylife cycle. It is virtually certain that the more compre-hensive and detailed models of the future will havesome sort of hybrid structure along the lines of theAthale and Wakeland models.

Obtaining Essential Biological InsightsThe preceding examples of biomedical ABMs havefocused on implementing known mechanistic detailsobtained from wet lab experiments into in silicoanalogues of their biological reference systems. Thisapproach represents a dynamic representation ofknowledge that allows the examination of theresulting ABM behaviors as a means of evaluating theveracity of the underlying hypotheses/mental model.This type of application of agent-based modeling isgenerally carried out with an explicit translational goalin mind (be it translation across organizational scales,domains of expertise, or to the bedside). However,there is a slightly different use of agent-based modelingthat is directed at obtaining insights into the functionand actions of biological systems. The ability of ABMs

to generate complex behavior from relatively simplerules allows a different tack to be applied to theperceived need for continually increasing mechanisticdetail. By constructing ABMs with rules that representcore and basic functions, they may aid in identifyingfunctional modules that represent yet another scale oforganization for cellular behavior.

When this approach is used in the study ofintracellular and subcellular processes, these modulesmay lead to the identification of potential points ofbiomolecular control. There are strong suggestionsthat while the functional modules may be internallyrobust, they have finely regulated control points attheir articulations with other modules, essentiallycreating ‘bottlenecks’ that regulate the outcome ofsignaling events. This phenomenon has been variablydescribed as the ‘hourglass’ or ‘bowtie’ structureof dynamic biological systems.44,45 One example ofthis can be seen in an ABM of bilayer membranestructure and behavior.46 This ABM uses a series ofrules, similar to that of the canonical bird flockingABM, to represent the interactions present among theconstituent molecules of a semifluid cell membrane.This model is particularly notable for being ableto reproduce the self-organization and subsequentproperties of a cell membrane while ignoring theissue of polarity of the constituent molecules. Evenif the balance between attraction and repulsion withina molecule is represented uniformly, the tendencyto self-organize into spheroids and micelles stillpredominates.

The work of the Hunt laboratory presentsthis approach more formally and explicitly intheir models of tumor spheroid growth.13,32 Again,by concentrating on representing key functionalbehaviors as agent rules, Hunt and colleagues wereable to identify nine key behaviors (which they term‘axioms’) that form the necessary and sufficient set ofbehaviors for the development of tumor topologies(Figure 3). The identification of these ‘essential’behavioral sets may provide a means of refining thecharacterization of those mechanisms to be targetedin the design of therapeutic interventions.

Cancer research has also provided examples ofthis use of agent-based modeling. One recent exampleis the work of Abbott et al.47I in which the authorsimplemented an ABM of Hanahan and Weinberg’ssix ‘Hallmarks of Cancer,’48 and predicted the mostlikely pathways for tumor formation, as well asother insights that have the potential to lead to newtargets for clinical work. Another example is thework of the Maley laboratory that has examinedthe evolutionary and ecological aspects of tumorgrowth.49 By using an ABM of cellular differentiation,



they have identified cellular characteristics andproperties that are susceptible to evolutionary forcesleading to neoplastic dynamics.50 These findingshave implications with respect to the robustnessof certain cellular properties to modification, andthus suggest mechanisms of tumor cell resistance tochemotherapeutic modalities.

Despite these successes, various challengesremain to the widespread use of ABM in the transla-tional process. We discuss some of these limitationsbelow.

CHALLENGES TO THE USE OFAGENT-BASED MODELINGIn the foregoing discussion, we have highlightedthe advantages of and advances made with ABM.However, this modeling framework is not withoutits limitations. A key limitation of ABM is sharedwith all computational modeling methods: the qualityand reliability of the models are directly relatedto the reliability of the underlying assumptions ofthe model and the quality of their implementationduring construction of the model. Given the shiftingnature and intrinsic incompleteness of biomedicalknowledge, the primary means of addressing thisissue in the agent-based modeling community hasbeen to emphasize transparency with respect tounderlying assumptions. Toward this end, the agent-based modeling community has made significantefforts to provide detailed and explicit references tothe knowledge base underlying their models, and therelationship between the in silico implementation andtheir biological referents (the following list is notcomprehensive but provide excellent examples of thisprocess11,14,27,38,51–53).

One perceived shortcoming of agent-basedmodeling is the difficulty in applying formal analysisto the relationship between the agent rules and thebehavior of the system. Because of the combinedstochastic behavior of agents, the very feature thatallows ABM to transcend the epistemological bound-aries inherent to multiscale structure, it is extremelydifficult to systematically analyze how each part orparameter of an ABM simulation affects the outputor behavior of the ABM. This stands in contrast toequation-based modeling, for which analytical taskssuch as parameter sensitivity analysis, bifurcationanalysis, and behavior-space determination can becarried out using well-established procedures. Thedifficulty in carrying out similar procedures for ABMscan make researchers with a more traditional math-ematical modeling background apprehensive in theirability to evaluate an ABM. This apprehension can be

addressed by viewing ABMs as objects more akin towet lab experimental platforms rather than more tra-ditional, equation-based mathematical models. Usingthe principles of pattern-oriented analysis, in whichcorresponding patterns of dynamic behavior are usedto relate the computational ABM to its real-worldreferent, ABM can be measured and analyzed in thesame fashion as wet lab systems or organisms.10 Inthis context, the stochastic nature of ABMs actuallyaids in recreating the robustness of dynamic behaviorseen in complex systems, perhaps a more realisticproperty than the ‘brittleness’ often seen in equation-based models. The downside of this characteristic,however, is that modelers using ABM must rely uponbrute force computation in order to generate densedata sets amenable to statistical analysis.

On one hand, the in silico nature of ABMprovides an advantage over wet lab approachesbecause of their ability to generate these dense andextensive data sets; electrons are much cheaper andfaster than cells, animals, and reagents. On the otherhand, there is a high computational cost associatedwith running ABM as compared to equation-basedmodels. Current freely available ABM platforms runas emulated parallel processing systems based on asingle-threaded central processing unit. The executionof an ABM in these settings requires multiple iteratedcomputations as each discrete event is carried out,many more than for equation-based simulations.This iterative process results in significantly greatercomputational demands, and these increased demandsconstrain the size of ABM implementations thatcan be run in the typical academic setting. Thenatural solution to this bottleneck is to implementABMs on the newest generation of high performancesupercomputers and other distributed computingplatforms. However, there are intrinsic properties ofABMs, namely related to spatial continuity and agent-to-agent communication, that challenge the abilityto implement ABM in a distributed environment.To address this bottleneck, researchers have startedto explore novel computing environments for agent-based modeling, such as using Graphical ProcessingUnits (GPUs).54 In addition, work has been initiatedon parallelizing aspects of the SPARK code (A.Solovyev, M. Mikheev, Q. Mi, and Y. Vodovotz,unpublished observations). It should be noted thatthere are nontrivial modeling issues associated withparallel implementation of ABM, aside from thecomputer science challenges. These primarily derivefrom the selection of the scale of process tobe distributed across the multiple processors andthe consequences that decision has with respectto the mapping of the simulation behavior back



to the biological referent. For instance, thus far theapproaches listed above have not explored the distri-bution of a single agent’s execution across multipleprocessors, and have opted for the ‘cleaner’ distri-bution scale of expanding the interaction space andimplementing the agents on the same level.

Other methods are also being implemented inorder to increase the feasibility of large ABM-basedsimulations. Several groups have pursued work relatedto the development of ‘hybrid’ model systems, inwhich differential equations are used to model thoseaspects of a system in which mean-field approxi-mations are valid, while ABMs are utilized wherespatial heterogeneity and its effects are significant.41,43

Additionally, methods are being developed to algorith-mically increase the efficiency of the evaluation andanalysis of complex, multiscale models.55

CONCLUSION

The coming era of translationally useful computa-tional simulations raises the potential for the devel-opment of highly predictive, personalized simulationsthat hold the promise of streamlining the design oftherapies, simulating the clinical application of thesetherapies in population studies (clinical trials), andpredicting the effects of drugs on individuals. Agent-based modeling, developed originally for the socialsciences, has been central to this rapidly expandingfield of translational systems biology. The majority of

algorithms and methods utilized for dynamical sim-ulations have been in the realm of equation-basedmodeling. However, developments in computationalmethods specific to agent-based modeling are likelyto greatly increase the utility of this user-friendlyapproach to modeling complex biological systems.

The agent-based paradigm, incorporating encap-sulation, modularity and parallelism, can provide anoverarching architecture for the computational rep-resentation of biological systems. By decreasing thethreshold for the general researcher to utilize com-putational modeling, agent-based modeling allowsinvestigators to ‘see’ the consequences of a particu-lar hypothesis-structure/conceptual model, such thatthe mechanistic consequences of each hypothesis canbe observed and evaluated. This type of dynamicknowledge representation enables the instantiation of‘thought experiments,’ i.e., trying out possible alter-native solutions, so long as these hypotheses andassumptions are made explicit. This concept drawsupon the experience in the Artificial Life communityby creating ‘alternative worlds’ driven by these pro-posed rules. These models can aid in the scientificprocess by providing a transparent framework for thistype of speculation, which can then be used as jump-ing off points for the planning and design of furtherwet laboratory experiments and measurements. It ishoped that the increasing use of this type of knowl-edge representation and communication will fosterthe further development of ‘virtual laboratories’ andin silico investigations.

ACKNOWLEDGEMENTS

This work was supported by National Institutes of Health (NIH) grants R33-HL-089082, P50-GM-53789, R01-HL080926, and R01-DC-008290; by National Institute on Disability Rehabilitation Research (NIDRR) grantH133E070024; by National Science Foundation (NSF) grant 0829864; by the Commonwealth of Pennsylvania;by a the Pittsburgh Tissue Engineering Initiative; and by a Shared University Research award from IBM, Inc.

REFERENCES

1. An G. Concepts for developing a collaborative in sil-ico model of the acute inflammatory response usingagent-based modeling. J Crit Care 2006, 21:105–110;discussion 110–101.

2. Abbott R. The reductionist blind spot. Complexity2009, 14:5, 10–22.

3. An G, Faeder J, Vodovotz Y. Translational systemsbiology: introduction of an engineering approach tothe pathophysiology of the burn patient. J Burn CareRes 2008, 29:277–285.

4. Vodovotz Y, Csete M, Bartels J, Chang S, An G.Translational systems biology of inflammation. PLoSComput Biol 2008, 4:e1000014.

5. An G. Introduction of an agent-based multi-scale mod-ular architecture for dynamic knowledge representationof acute inflammation. Theor Biol Med Model 2008,5:11.

6. Fisher J, Henzinger TA. Executable cell biology. NatBiotechnol 2007, 25:1239–1249.



7. Hunt CA, Ropella GE, Park S, Engelberg J. Dichotomiesbetween computational and mathematical models. NatBiotechnol 2008, 26:737–738; author reply 738–739.

8. Bankes SC. Agent-based modeling: a revolution?. ProcNatl Acad Sci U S A 2002, 99(suppl 3):7199–7200.

9. Bonabeau E. Agent-based modeling: methods and tech-niques for simulating human systems. Proc Natl AcadSci U S A 2002, 99(suppl 3):7280–7287.

10. Grimm V, Revilla E, Berger U, Mooij W, Railsback S,et al. Pattern-oriented modeling of agent-based com-plex systems: lessons from ecology. Science 2005,310:987–991.

11. An G. Agent-based computer simulation and sirs: build-ing a bridge between basic science and clinical trials.Shock 2001, 16:266–273.

12. Zhang L, Athale CA, Deisboeck TS. Development of athree-dimensional multiscale agent-based tumor model:simulating gene-protein interaction profiles, cell pheno-types and multicellular patterns in brain cancer. J TheorBiol 2007, 244:96–107.

13. Engelberg JA, Ropella GE, Hunt CA. Essential oper-ating principles for tumor spheroid growth. BMC SystBiol 2008, 2:110.

14. Hunt CA, Ropella GE, Yan L, Hung DY, RobertsMS. Physiologically based synthetic models of hep-atic disposition. J Pharmacokinet Pharmacodyn 2006,33:737–772.

15. Reynolds CW. Flocks, herds, and schools: a dis-tributed behavioral model in computer graphics. Com-put Graphics 1987, 21:25–34 (SIGGRAPH ’87 Con-ference Proceedings).

16. Lipniacki T, Paszek P, Brasier AR, Luxon BA, Kim-mel M. Stochastic regulation in early immune response.Biophys J 2006, 90:725–742.

17. Lipniacki T, Paszek P, Marciniak-Czochra A, BrasierAR, Kimmel M. Transcriptional stochasticity in geneexpression. J Theor Biol 2006, 238:348–367.

18. Vodovotz Y, Clermont G, Hunt CA, Lefering R, BartelsJ, et al. Evidence-based modeling of critical illness: aninitial consensus from the Society for Complexity inAcute Illness. J Crit Care 2007, 22:77–84.

19. Wilensky U, Rand W. An introduction to agent-basedmodeling: modeling natural, social and engineered com-plex systems with NetLogo. Cambridge, MA: MITPress; 2009.

20. Bailey AM, Lawrence MB, Shang H, Katz AJ, PeirceSM. Agent-based model of therapeutic adipose-derivedstromal cell trafficking during ischemia predicts abil-ity to roll on P-selectin. PLoS Comput Biol 2009,5:e1000294.

21. Railsback SF, Lytinen SL, Jackson SK. Agent-basedsimulation platforms: review and development recom-mendations. Simulation 2006, 82:609–623.

22. Folcik VA, An GC, Orosz CG. The Basic Immune Sim-ulator: an agent-based model to study the interactions

between innate and adaptive immunity. Theor Biol MedModel 2007, 4:39.

23. Vodovotz Y, Constantine G, Rubin J, Csete M, VoitEO, et al. Mechanistic simulations of inflammation:Current state and future prospects. Math Biosci 2009,217:1–10.

24. Ridgway D, Broderick G, Lopez-Campistrous A,Ru’aini M, Winter P, et al. Coarse-grained molec-ular simulation of diffusion and reaction kineticsin a crowded virtual cytoplasm. Biophys J 2008,94:3748–3759.

25. Pogson M, Smallwood R, Qwarnstrom E, HolcombeM. Formal agent-based modelling of intracellular chem-ical interactions. Biosystems 2006, 85:37–45.

26. Pogson M, Holcombe M, Smallwood R, QwarnstromE. Introducing spatial information into predictive NF-kappaB modelling–an agent-based approach. PLoSONE 2008, 3:e2367.

27. An G. A model of TLR4 signaling and tolerance using aqualitative, particle-event-based method: introductionof spatially configured stochastic reaction chambers(SCSRC). Math Biosci 2009, 217:43–52.

28. Deisboeck TS, Berens ME, Kansal AR, TorquatoS, Stemmer-Rachamimov AO, et al. Pattern of self-organization in tumour systems: complex growthdynamics in a novel brain tumour spheroid model.Cell Prolif 2001, 34:115–134.

29. Thorne BC, Bailey AM, Peirce SM. Combining experi-ments with multi-cell agent-based modeling to studybiological tissue patterning. Brief Bioinform 2007,8:245–257.

30. Grant MR, Mostov KE, Tlsty TD, Hunt CA. Simulat-ing properties of in vitro epithelial cell morphogenesis.PLoS Comput Biol 2006, 2:e129.

31. Segovia-Juarez JL, Ganguli S, Kirschner D. Identifyingcontrol mechanisms of granuloma formation duringM. tuberculosis infection using an agent-based model.J Theor Biol 2004, 231:357–376.

32. Chen S, Ganguli S, Hunt CA. An agent-based compu-tational approach for representing aspects of in vitromulti-cellular tumor spheroid growth. Conf Proc IEEEEng Med Biol Soc 2004, 1:691–694.

33. Yu W, Fang X, Ewald A, Wong K, Hunt CA,et al. Formation of cysts by alveolar type II cellsin three-dimensional culture reveals a novel mecha-nism for epithelial morphogenesis. Mol Biol Cell 2007,18:1693–1700.

34. Grant MR, Kim SH, Hunt CA. Simulating in vitroepithelial morphogenesis in multiple environments.Comput Syst Bioinformatics: Life Sci Soc 2006,381–384.

35. Longo D, Peirce SM, Skalak TC, Davidson L, MarsdenM, et al. Multicellular computer simulation of morpho-genesis: blastocoel roof thinning and matrix assemblyin Xenopus laevis. Dev Biol 2004, 271:210–222.



36. Peirce SM, Van Gieson EJ, Skalak TC. Multicellularsimulation predicts microvascular patterning and insilico tissue assembly. Faseb J 2004, 18:731–733.

37. Walker DC, Hill G, Wood SM, Smallwood RH,Southgate J. Agent-based computational modelingof wounded epithelial cell monolayers. IEEE TransNanobioscience 2004, 3:153–163.

38. Mi Q, Riviere B, Clermont G, Steed DL, Vodovotz Y.Agent-based model of inflammation and wound heal-ing: insights into diabetic foot ulcer pathology andthe role of transforming growth factor-beta1. WoundRepair Regen 2007, 15:671–682.

39. Li NY, Verdolini K, Clermont G, Mi Q, RubinsteinEN, et al. A patient-specific in silico model of inflam-mation and healing tested in acute vocal fold injury.PLoS ONE 2008, 3:e2789.

40. Mansury Y, Diggory M, Deisboeck TS. Evolutionarygame theory in an agent-based brain tumor model:exploring the ‘Genotype-Phenotype’ link. J Theor Biol2006, 238:146–156.

41. Athale CA, Deisboeck TS. The effects of EGF-receptordensity on multiscale tumor growth patterns. J TheorBiol 2006, 238:771–779.

42. Athale C, Mansury Y, Deisboeck TS. Simulating theimpact of a molecular ‘decision-process’ on cellularphenotype and multicellular patterns in brain tumors.J Theor Biol 2005, 233:469–481.

43. Wakeland W, Macovsky L, An G. A hybrid simu-lation for studying the acute inflammatory response.Proceedings of the 2007 Spring Simulation Multicon-ference (Agent Directed Simulation Symposium) 2007,1:39–46.

44. Kitano H, Oda K. Robustness trade-offs and host-microbial symbiosis in the immune system. Mol SystBiol 2006, 2. doi:10.1038/msb410003.

45. Kitano H. Towards a theory of biological robustness.Mol Syst Biol 2007, 3:137. doi:10.1038/msb4100179.

46. Broderick G, Ru’aini M, Chan E, Ellison MJ. A life-likevirtual cell membrane using discrete automata. In SilicoBiol 2005, 5:163–178.

47. Abbott RG, Forrest S, Pienta KJ. Simulating the hall-marks of cancer. Artif Life 2006, 12:617–634.

48. Hanahan D, Weinberg RA. The hallmarks of cancer.Cell 2000, 100:57–70.

49. Merlo LM, Pepper JW, Reid BJ, Maley CC. Cancer asan evolutionary and ecological process. Nat Rev 2006,6:924–935.

50. Pepper JW, Sprouffske K, Maley CC. Animalcell differentiation patterns suppress somatic evo-lution. PLoS Comput Biol 2007, 3(12):e250.doi:10.1371/journal.pcbi.0030250.

51. Tang J, Ley KF, Hunt CA. Dynamics of in silico leuko-cyte rolling, activation, and adhesion. BMC Syst Biol2007, 1:14.

52. An G. In silico experiments of existing and hypothet-ical cytokine-directed clinical trials using agent-basedmodeling. Critical care medicine 2004, 32:2050–2060.

53. Bailey AM, Thorne BC, Peirce SM. Multi-cell agent-based simulation of the microvasculature to study thedynamics of circulating inflammatory cell trafficking.Ann Biomed Eng 2007, 35:916–936.

54. Richards RS, D’Souza R, Lysenko M, An G. Data-parallel techniques for agent-based tissue modeling ongraphical processing units. Design Engineering Tech-nical Conference and Computers and Information inEngineering Conference. New York, 2008.

55. Kevrekidis I, Gear C, Hummer G. Equation-free: thecomputer-aided analysis of complex multiscale systems.AIChE J 2004, 50:1346–1355.


Agent-based models in translational systems biology

Documents