Swarm-based Immune System Simulation andEvolution

Christian Jacob1,2

1 Dept. of Computer Science, Faculty of Science2 Dept. of Biochemistry & Molecular Biology, Faculty of Medicine,

University of Calgary, Calgary, Alberta, Canada cjacob@ucalgary.ca

Summary. We consider the human body as a well-orchestrated system of inter-acting swarms. Utilizing swarm intelligence techniques, we explore key aspects ofthe human immune system utilizing our latest virtual simulation and experimen-tation environment, IMMS:VIGO::3D. In our models, immune system cells and re-lated entities (viruses, bacteria, cytokines) are represented as virtual agents inside3-dimensional, decentralized and compartmentalized environments that representprimary and secondary lymphoid organs as well as vascular and lymphatic vessels.Specific immune system responses emerge as by-products from collective interactionsamong the involved simulated ‘agents’ and their environment. We demonstrate sim-ulation results for clonal selection and primary and secondary collective responsesafter viral infection, as well as the key response patterns encountered during bacterialinfection. Using an evolutionary optimization approach to adjust control parametersof the model and its simulated immune system agents is show by example which il-lustrates the potential of evolutionary techniques to help in building and adjustingcomputer-based models to in vivo systems.

1 Introduction

Computer-based tools and virtual simulations are changing the way of bio-logical research. Immunology is no exception. Computers become even morecapable of running large-scale models of complex biological systems. Recentadvancements in grid computing technologies make high-performance com-puter resources readily accessible to almost everybody [1]. Consequently, evenhighly sophisticated – and to a large extent still poorly understood – processessuch as the inner workings of immune system defense mechanisms can now betackled by agent-based models in combination with interactive visualizationcomponents. These agent models serve as an essential complement to model-ing approaches that are traditionally more abstract and purely mathematical[2, 3].

2 Christian Jacob,

Our Evolutionary & Swarm Design Laboratory is building and promotingagent-based models, with distributed simulation and visualization capabili-ties, utilizing swarm intelligence methodologies [4]. The combination of visualand intuitive user interfaces, in combination with the latest technology in vi-sualization (including 3D-immersive environments in CAVES) and distributedhigh-performance computing, makes our models more accessible to researchersin the life sciences community, who usually do not have any programmingbackground or any aspiration nor time to learn how to use a modeling en-vironment. Making simulation tools (almost) seamless to use for researchersand introducing such tools into classrooms in biology and medicine greatlyincreases the understanding of how useful computer-based simulations can bein order to explore and facilitate answers to research questions and, as a sideeffect, gain an appreciation of emergent effects resulting from orchestratedinteractions of ‘bio-agents’.

In this paper we present our latest version of a swarm-based simulationenvironment, which, we think, fulfills these criteria, and implements an inter-active virtual laboratory for the exploration of the interplay of human immunesystem agents and their resulting overall response patterns. The rest of thepaper is organized as follows. In Section 2 we give an overview of relatedsimulation and modeling approaches regarding immune system processes. Abiological perspective of the decentralized immune defenses is presented inSection 3. The key design aspects and main results of our IMMS:VIGO::3Dsimulation system are described in Section 4, where we also discuss simu-lation experiments for clonal selection, primary and secondary responses toviral infection, as well as reactions to bacterial infection. Section 5 describeshow genetic programming techniques can be used to either manually breedor automatically evolve simulation parameters of our immune system modelsfor specific fitness criteria. Finally, in Section 6, we conclude the paper witha summary of our work and suggestions for the necessary next steps towardsan encompassing immune system simulation environment.

2 Related and Previous Work

The immune system (IS) has been studied from a modeling perspective for along time. Early, more general approaches looked at the immune system inthe context of adaptive and learning systems [5, 6], with some connectionsto early artificial intelligence approaches [7]. Purely mathematical models,mainly based on differential equations, try to capture the overall behaviourpatterns and changes of concentrations during immune system responses [2, 3,8, 9, 10]. A more recent algebraic model of B and T cell interactions providesa formal basis to describe binding and mutual recognition, and can serve asa mathematical basis for further computational models, similar to formalismsfor artificial neural networks [11].

Swarm-based Immune System Simulation and Evolution 3

Agent-based computational approaches, in the form of cellular automata,introduced spacial aspects to immune system simulations [12]. In the con-text of clonal selection, the influence of different affinities among interactingfunctional units, which leads to self-organizing properties, was recognized andstudied through computational models [13, 14]. These models have been ex-panded into larger and more general simulation environments for various as-pects of the human immune system [15, 16]. There is also a large number ofmodeling approaches within specific areas in the context of immune system-related processes, such as for HIV/AIDS [17]. An excellent overview of thesemodeling strategies can be found in [18].

Most current methods consider immune response processes as emergentphenomena in complex adaptive system [10], where agent-based models playa more and more dominant role [19, 20], even in the broader application do-main of bio-molecular and chemical interaction models [21]. We see the mostpromising potential in agent models that incorporate swarm intelligence tech-niques [4, 22], as this results in more accurate and realistic models, in partic-ular when spacial aspects play a key role in defining patterns of interaction,in understanding their emergent properties, and in helping to shed some lighton the inner workings of complexity as, for example, displayed by the immunesystem. Biological systems inherently operate in a 3-dimensional world. There-fore, we have focused our efforts on building swarm-based, 3-D simulations ofbiological systems which exhibit a high degree of self-organization, triggeredby relatively simple interactions of a large number of agents of different types.The immune system is just one example that allows for this bottom-up mod-eling approach. Other models include the study of chemotaxis within a colonyof evolving bacteria [23, 24], the simulation of transcription, translation, andspecific gene regulatory processes within the lactose operon [25, 26], as wellas studies of affinity and cooperation among gene regulatory agents for the λswitch in E. coli [27].

3 The Decentralized Defenses of Immunity

One aspect that makes the human immune system particularly interesting—but more challenging from a modeling perspective—is its vastly decentralizedarrangement. Tissue and organs of the lymphatic system are widely spreadthroughout the body, which provides good coverage against any infectiousagents that might enter the body at almost any location. Even the two keyplayers responsible for specific immunity originate from different locationswithin the body: T cells come from the thymus, whereas B cells are madein the bone marrow. The lymphocytes then travel through the blood streamto secondary lymphoid organs: the lymph nodes, spleen, and tonsils. Withinthese organs, B and T cells are rather tightly packed, but can still movearound freely, which makes them easier to model as agents interacting in a3-D simulation space.

4 Christian Jacob,

Lymph nodes can then be considered the primary locations of interac-tions among T cells, activated by antigens. T cells, in turn, activate B cells,which evolve into memory B cells and antibody-producing plasma B cells.Both types of activated lymphocytes will subsequently enter the lymphaticsystem, from where they eventually return to the blood stream. This enablesthe immune system to spread its activated agents widely through the body.Finally, the lymphocytes return to other lymph nodes, where they can recruitfurther agents or trigger subsequent responses. Hence, B and T cells as wellas other immune system agents (antibodies, cytokines, dendritic cells, antigenpresenting cells, etc.) are in a constant flow between different locations in thehuman body [28].

Fig. 1. The decentralized defenses of immunity. Three compartmental modules, thatexhibit distinct but interconnected functionalities within the human immune system,are implemented in our IMMS:VIGO::3D simulation environment: (1) tissue, (2)blood vessels, and (3) lymph nodes.

4 Simulating Decentralized Immune Responses

Our overall goal is to build a whole body simulation of the immune system(Fig. 1). This, of course, does not only require a large amount of computingresources, but also requires a modular and hierarchical design of the simulationframework. Modelers – i.e., immunologists as well as researchers and studentsin health sciences – should be able to look at the simulated immune system atdifferent levels of detail. The whole body simulation will not be as fine grained

Swarm-based Immune System Simulation and Evolution 5

as when looking at the interactions within a lymph node or at the intersectionbetween the lymphatic and vascular system.

In our current implementation we have incorporated three distinct, butinterconnected sites within the human body that are related to the immunesystem:

• Lymph Nodes: Within a lymph node section we incorporate adaptiveimmune system processes during clonal selection, in response to viral anti-gens entering the lymph node. Different types of B cell strands can bedefined. In case of a high degree of matching with an antigen, rapid pro-liferation is triggered.

• Tissue: Within a small section of tissue we model the immune systemprocesses during primary and secondary response reactions among viruses(with their associated antigen components), tissue cells, dendritic cells,helper T and killer T cells, memory and plasma B cells (with their associ-ated antibodies), and macrophages.

• Blood Vessel-Tissue Interfaces: At the interface between blood ves-sels and tissue, we simulate red blood cells moving within a section of ablood vessel, lined with endothelial cells, which can produce selectin andintercellular adhesion molecules (ICAMs). This causes neutrophils to startrolling along the vessel wall and exit the blood stream into the tissue area.Any bacterium within the tissue is subsequently attacked by a neutrophil.During ingestion of a bacterium by a macrophage, tumor necrosis factor(TNF) is secreted and the bacterium releases lipopolysaccharides (LPS)from its surface. In turn, TNF triggers selectin production in endothelialcells, whereas LPS induces endothelial cells to produce ICAM.

The following sections explain our model in more detail with respect toclonal selection as well as primary and secondary responses within a lymphnode area and a tissue region (Section 4.1). The IS processes triggered duringa bacterial infection within the interface area between a blood vessel and tissueis described in Section 4.2.

4.1 Simulated Viral Infection

Figure 2 gives an overview of the immune system agents and their interactionpatterns in our model. Each agent is represented by a specific, 3-dimensionalshape, which are also used in the (optional) visual representation of the agentsduring a simulation experiment. We demonstrate one experiment to show atypical simulation sequence.

Clonal Selection within a Lymph Node:

In this experiment, we first focus our attention on a selected lymph node inorder to observe the IS agent reactions after a virus enters the lymph node

6 Christian Jacob,

Memory B-Cell

B-Cell

PlasmaB-Cell

Dendritic Cell

Helper T-Cell

Macrophage

Killer T-Cell

Antibodies Virus

Antigen

Tissue Cellopsonizes infects

destroysinfected

cell

activatesactivates

recruits recruits

eaten by eaten by

recruits recruits

secrete

specializes

secretes

specializes

2nd exposure to an antigen stimulates Memory B-Cells to produce corresponding Plasma B-Cells Humoral Immunity

Cell-Mediated Immunity

Fig. 2. Interactions of immune system agents triggered by viral infection: A virus isusually identified by its antigens, which alert both dendritic cells and macrophages toingest the viruses. Both actions lead to recruitment of further IS cells. Dendritic cellsrecruit B cells, which – in particular when activated by helper-T cells – replicate asmemory B cells or proliferate into plasma B cells, which in turn release antibodies toopsonize the virus. On the other hand, macrophages with an engulfed virus stimulatean increase in the proliferation of both helper and killer T cells, which are the keyplayers in cell-mediated immunity and destroy virus-infected tissue cells to preventany further spreading of the virus.

area (cf. Fig. 1). Initially, 50 B cells as well as 20 helper-T cells of 8 differenttypes (signatures) are present. Figure 3f shows that there is a fairly eveninitial distribution of the different strands of B and T cells. Around time stept = 14.6, dendritic cells enter the lymph node and present a single type ofviral antigen (Fig. 3b), which stimulates a nearby helper-T cell and causes amatching B cell (following the Celada-Seiden affinity model [12]) to replicate.Soon after (t = 57.1), a significantly larger population of matching B cellsproliferates the lymph node area (Fig. 3c), where B cells have already startedto emit antibodies. In Fig. 3f the concentration of these fast proliferating Bcells is represented by the green plot. At time point t = 225.0, memory Bcells of the matching strand have become more common. Around t = 256.4,the same virus is introduced into the lymph node again. Now it is mainly the

Swarm-based Immune System Simulation and Evolution 7

memory B cells that trigger the secondary response and replication of plasmaB cells which secrete antibodies (compare the increase of the matching B cellconcentration (green) towards the last third of the graph in Fig. 3f).

Primary and Secondary Response in Tissue:

At the same time, while the simulation of the interactions within the lymphnode are running, a concurrent, second simulation models the response pro-cesses in a selected tissue area (cf. Fig. 1). Circulation of IS agents is imple-mented by a communication channel between lymph node and tissue areas.Within the tissue simulation space (Fig. 4a), we start with 10 dendritic cells,5 killer-T cells, 5 helper-T cells, 5 macrophages, 60 tissue cells and 5 copies ofthe same virus introduced into the lymph node as described above. In Fig. 4ba cell has been infected by the virus and antibodies (from the lymph node)start entering the tissue area. Figure 4c shows a close-up of the importantagents: one virus is visible inside an infected cell, another virus has dockedonto the surface of a tissue cell and is about to enter it. A third virus hasalready been opsonized by an attached antibody. Now macrophages will startto engulf opsonized viruses and more macrophages are recruited in large num-bers (Fig. 4d). This triggers an analogous spike in the number of killer-T andhelper-T cells (also compare Fig. 5). The increase in killer-T cells makes itmore likely for these cells to collide with an infected tissue cell and initiate itsapoptosis.

After about 120 time steps, the infection has been fought off, with no moreviruses or antigens remaining in the system (Fig. 5). The concentrations ofT cells and macrophages return to their initial levels. At t = 150.0, the samevirus is reinserted into the system. Memory B cells inside the lymph nodecreate an influx of plasma B cells almost immediately. Due to the increasedamount of antibodies emitted, the infection is stopped within a much shortertime interval. As a result, the infection is stopped within a much shorter timeinterval, due to the increased amount of antibodies. Cell-mediated immunityreactions do start faster as well, but are not as intense as during the firstresponse since the infection is eliminated more quickly. Consequently, T cellsand macrophage concentrations can remain at a lower level.

4.2 Simulated Bacterial Infection

Bacteria within the tissue multiply; their waste products, produced from alarge concentration of bacteria, can be damaging to the human body. There-fore it is important that the immune system kills off bacterial invaders beforethis critical concentration is reached. The following experiment demonstratesimmune system response processes during bacterial infection. The key playersand their interactions are outlined in Fig. 6. As this involves not only bacte-ria and macrophages but also neutrophils that enter tissue from the vascular

8 Christian Jacob,

(a) t = 3.2 (b) t= 14.6

(c) t = 57.1 (d) t = 225.0

0 50 100 150 200 250 300

0

20

40

60

80

B!Cells

0 50 100 150 200 250 300

0

20

40

60

80

B!Cells

(e) t = 256.4 (f) Concentration of B cell strands

Fig. 3. Interactions in a Lymph Node after a viral infection: (a)-(e) Screen cap-tures (with time point labels) of the graphical simulation interface during clonalselection and primary and secondary response to a virus. The virtual cameras arepointed at a lymph node, in which 8 different strands of B cells are present. (f) Thechange in concentration of all B cells (brown filled plot) and per strand. The virusthat most closely matches one of the B cell strands triggers its increased proliferation(green filled plot). The concentrations of all other strands remain low (line plots atthe bottom).

Swarm-based Immune System Simulation and Evolution 9

(a) t = 4.4 (b) t= 25.1

(c) t = 40.4 (d) t = 73.6

(e) t = 225.0 (f) t = 268.8

Fig. 4. Interactions in a Tissue Area after a viral infection: Screen captures of thegraphical simulation interface during clonal selection and primary and secondaryresponse after viral infection. The virtual cameras are pointed at a tissue regionclose to a blood vessel.

10 Christian Jacob,

50 100 150 200 250t

10

20

30

40

50

60

!C! Tissue Cells

50 100 150 200 250t

10

20

30

40

50

60

!C! Tissue Cells

50 100 150 200 250t

5

10

15

20

25

!V! Viruses !infectious"

50 100 150 200 250t

5

10

15

20

25

!V! Viruses !infectious"

50 100 150 200 250t

2

4

6

8

10

!D! Dendritic Cells

50 100 150 200 250t

2

4

6

8

10

!D! Dendritic Cells

50 100 150 200 250t

10

20

30

40

50

!Ag! Antigens

50 100 150 200 250t

10

20

30

40

50

!Ag! Antigens

50 100 150 200 250t

2.5

5

7.5

10

12.5

15

!H! Helper T!Cells

50 100 150 200 250t

2.5

5

7.5

10

12.5

15

!H! Helper T!Cells

50 100 150 200 250t

50

100

150

200

250

300

!Ab! Antibodies

50 100 150 200 250t

50

100

150

200

250

300

!Ab! Antibodies

50 100 150 200 250t

2.5

5

7.5

10

12.5

15

!KT! Killer T!Cells

50 100 150 200 250t

2.5

5

7.5

10

12.5

15

!KT! Killer T!Cells

50 100 150 200 250t

5

10

15

20

25

!M! Macrophages

50 100 150 200 250t

5

10

15

20

25

!M! Macrophages

Fig. 5. Evolution of IS agent concentrations during the primary and secondaryresponses in a tissue area.

Swarm-based Immune System Simulation and Evolution 11

Bacteria

Hunting Neutrophil

Macrophage

LPS

TNF

releaseson death

EndothelialCell

collides with

collides with

kills

kills

ICAMEndothelial

Cell

SelectinEndothelial

Cell

Sliding Neutrophil

Neutrophil

and becomes

and becomes

activates

activates

exits blood stream and becomes

becomes

Fig. 6. Bacterial Infection: A summary of the interaction network between bacteria,macrophages, neutrophils, and endothelial cells that line the blood vessel.

(a) t = 14.0 (b) t = 17.6

(c) t = 37.2 (d) t = 71.7

Fig. 7. Fighting Bacterial Infection: (a) macrophages attacking bacteria, (b) en-dothelial cells, neutrophils and red blood cells inside the blood vessel, (c) neutrophils(blue) on their hunt for bacteria, (d) all bacteria have been eliminated.

12 Christian Jacob,

system, the simulation space comprises a segment of a blood vessel (Fig. 7).The tissue-vessel interface area is initialized with tissue cells, B cells, helper-Tcells, macrophages, and a number of bacteria acting as infectors. The bloodvessel, lined with endothelial cells, contains red blood cells and neutrophils.

Macrophages that engulf bacteria release TNF (tumor necrosis factor)while lipopolysaccharides (LPS), which are major structural components ofGram-negative bacterial cell walls, are released into the tissue area (Fig. 7a).Once endothelial cells get in contact with TNS or LPS, they release selectinor intercellular adhesion molecules (ICAMs), respectively (Fig. 7b). When aneutrophil collides with an endothelial cell which produces selectin, it willstart to roll along the interior surface of the blood vessel. A neutrophil rollingalong an ICAM-producing endothelial cell will exit the blood stream andhead into the tissue area. Once in the tissue area, neutrophils—together withmacrophages— act as complementary hunters of bacteria (Fig. 7c). Noticethe high number of activated endothelial cells in the blood vessel wall. A bac-terium colliding with a neutrophil is engulfed and consumed, while LPS andTNF are again released into the system. Finally, all bacteria have been elim-inated and the number of activated endothelial cells is decreased (Fig. 7d).Neutrophils will soon disappear since the system has recovered from the bac-terial infection.

5 Evolutionary Adaptation of Immune System Models

In this section we describe evolutionary experiments to adjust simulation pa-rameters of our immune system models. This section will be at least an addi-tional five pages.

6 Conclusion and Future Research

The IMMS:VIGO::3D simulation environment is currently used as a teachingtool in biology, medical, and computer science undergraduate and graduateclasses. Due to its visual interface and the ability to specify many simulationcontrol parameters through configuration files, it serves both as an educa-tional device as well as an exploration tool for researchers in the life sciences.Students seem to gain a more ‘memorable’ understanding of different aspectsof immune system processes. Although visualizations can also be misleading,they usually help in grasping essential concepts, in particular in the case of anorchestrated system of a multitude of agents. Consequently, from our experi-ence, the visualization component is important for a proper understanding ofemergent processes resulting from the interplay of a relatively large number ofagents of different types with simple but specific local interaction rules. Gain-ing a proper understanding and ‘intuition’ about emergent properties as in

Swarm-based Immune System Simulation and Evolution 13

Fig. 8. User interface for interactive evolution of immune system model parameters.

the immune system plays a key role in building today’s biologically accuratecomputer simulations.

Of course, our current version does not even come close to the actualnumbers of interacting IS agents (e.g., billions of B cells within a small lymphnode section). However, according to our experience, key effects within anagent-based interaction system can already be observed with much smallernumbers. Usually, only a ‘critical mass’ is needed. This is certainly an areathat requires further investigation, which we currently focus on. Using evolu-tionary computation techniques, we also explore the effects of different controlparameter settings, as well as how changes in the set of agent interaction rulesinfluence the overall system behaviour. Being able to easily change agent in-teraction rules and the types of agents makes models of complex adaptivesystems useful for large-scale scientific exploration.

Currently, we only have incorporated some of the earlier and basic theoriesof how immune system processes might work. Now that we have a flexibleand powerful simulation infrastructure in place, calibrating and validatingour models as well as including more of the recently proposed models is oneof our next steps. We are also expanding our simulations to demonstrate (andhelp students to investigate) why the generation of effective vaccines is difficultand how spontaneous auto-immunity emerges.

Up-to-date details about our latest immune system model and other agent-based simulation examples, which are investigated in our Evolutionary &Swarm Design Lab can be found at: http://www.swarm-design.org.

14 Christian Jacob,

Acknowledgements

Financial support for this research is provided by NSERC, the Natural Sci-ences and Engineering Research Council of Canada.

We thank Ian Burleigh for creating VIGO::3D, a C++ library for multi-agent simulation and visualization in 3D space [29], on which the IMMS:VIGO::3Dsystem is built. VIGO is an open source project available at:

http://sourceforge.net/projects/vigo.

References

1. Przybyla, D., Miller, K., Pegah, M.: A holistic approach to high-performancecomputing: xgrid experience. In ACM, ed.: Proceedings of the 32nd annualACM SIGUCCS conference on User services. (2004) 119–124

2. Salzberg, S., Searls, D., Kasif, S., eds.: Computational Methods in MolecularBiology. Volume 32 of New Comprehensive Biochemistry. Elsevier, Amsterdam(1998)

3. Bower, J.M., Bolouri, H., eds.: Computational Modeling of Genetic and Bio-chemical Networks. MIT Press, Cambridge, MA (2001)

4. Bonabeau, E., Dorigo, M., Theraulaz, G.: Swarm Intelligence: From Naturalto Artificial Systems. Santa Fe Institute Studies in the Sciences of Complexity.Oxford University Press, New York (1999)

5. Farmer, J.D., Packard, N.H.: The immune system, adaptation, and machinelearning. Physica D 22 (1986) 187–204

6. Bagley, R.J., Farmer, J.D., Kauffman, S.A., Packard, N.H., Perelson, A.S., Stad-nyk, I.M.: Modeling adaptive biological systems. BioSystems 23 (1989) 113–138

7. Rossler, O., Lutz, R.: A decomposable continuous immune network. BioSystems11 (1979) 281–285

8. Hanegraaff, W.: Simulating the immune system. Master’s thesis, Departmentof Computational Science, University of Amsterdam, Amsterdam, The Nether-lands (2001)

9. Bezzi, M., Celada, F., Ruffo, S., Seiden, P.E.: The transition between immuneand disease states in a cellular automaton model of clonal immune response.Physica A 245 (1997) 145 – 163

10. Tay, J.C., Jhavar, A.: Cafiss: a complex adaptive framework for immune systemsimulation. In: SAC ’05: Proceedings of the 2005 ACM symposium on Appliedcomputing, New York, NY, USA, ACM Press (2005) 158–164

11. Tarakanov, A., Dasgupta, D.: A formal model of an artificial immune system.BioSystems 55 (2000) 151–158

12. Celada, F., Seiden, P.E.: A computer model of cellular interactions in the im-mune system. Immunology Today 13(2) (1992) 56–62

13. Celada, F., Seiden, P.E.: Affinity maturation and hypermutation in a simulationof the humoral immune response. European Journal of Immunology 26 (1996)1350–1358

14. Atamas, S.P.: Self-organization in computer simulated selective systems. BioSys-tems 39 (1996) 143–151

Swarm-based Immune System Simulation and Evolution 15

15. Kleinstein, S.H., Seiden, P.E.: Simulating the immune system. Computing inScience & Engineering (July/August) (2000) 69–77

16. Puzone, R., Kohler, B., Seiden, P., Celada, F.: Immsim, a flexible model for inmachina experiments on immune system responses. Future Generation Com-puter Systems 18(7) (2002) 961–972

17. Guo, Z., Han, H.K., Tay, J.C.: Sufficiency verification of hiv-1 pathogenesisbased on multi-agent simulation. In: GECCO ’05: Proceedings of the 2005conference on Genetic and evolutionary computation, New York, NY, USA,ACM Press (2005) 305–312

18. Guo, Z., Tay, J.C.: A comparative study on modeling strategies for immune sys-tem dynamics under hiv-1 infection. In Jacob, C., Pilat, M., Bentley, P., Timmis,J., eds.: Artificial Immune Systems, 4th International Conference, ICARIS 2005.Volume LNCS 3627., Springer (2005) 220–233

19. Johnson, S.: Emergence: The Connected Lives of Ants, Brains, Cities, andSoftware. Scribner, New York (2001)

20. Wolfram, S.: A New Kind of Science. Wolfram Media, Champaign, IL (2002)21. Pogson, M., Smallwood, R., Qwarnstrom, E., Holcombe, M.: Formal agent-based

modelling of intracellular chemical interactions. BioSystems (2006)22. Camazine, S., Deneubourg, J.L., Franks, N.R., Sneyd, J., Theraulaz, G.,

Bonabeau, E.: Self-Organization in Biological Systems. Princeton Studies inComplexity. Princeton University Press, Princeton (2003)

23. Hoar, R., Penner, J., Jacob, C.: Transcription and evolution of a virtual bacteriaculture. In: Congress on Evolutionary Computation, Canberra, Australia, IEEEPress (2003)

24. Penner, J., Hoar, R., Jacob, C.: Bacterial chemotaxis in silico. In: ACAL 2003,First Australian Conference on Artificial Life, Canberra, Australia (2003)

25. Burleigh, I., Suen, G., Jacob, C.: Dna in action! a 3d swarm-based modelof a gene regulatory system. In: ACAL 2003, First Australian Conference onArtificial Life, Canberra, Australia (2003)

26. Jacob, C., Burleigh, I.: Biomolecular swarms: An agent-based model of thelactose operon. Natural Computing (2004) (in print).

27. Jacob, C., Barbasiewicz, A., Tsui, G.: Swarms and genes: Exploring λ-switchgene regulation through swarm intelligence. In: Congress on Evolutionary Com-putation, Vancouver, BC, Canada, IEEE Press (2006)

28. Nossal, G.J.: Life, death and the immune system. Scientific American (1993)53–62

29. Burleigh, I.G.: A journey to the center of the cell. Master’s thesis, Departmentof Computer Science, University of Calgary, Calgary, Canada (2004)