Human Immune System Simulation: A Survey of Current Approaches

John Daigle
Georgia State University
PDF

1 Introduction

Immune system simulation is a difficult task. The first problem that a simulator encounters is the size of a typical response, around 10¹² cells, each subtly different, are involved in a typical immune response. The subtly difficult part is important, because it limits the applicability of techniques such as ordinary differential equations for simulating accurate immune responses.[30] An alternate simulation approach is the use of Complex Adaptive Systems simulation.[70]

In either case, the problem is complex. With a large number of computations to be run, it makes sense to expect parallel or distributed computing to be useful.

This paper will be presented in several sections. Section 2 contains a brief introduction to the immune system itself, and the historical context of biological simulation. Section 3 discusses the history and development of discrete even and agent based simulation. Section 4 examines approaches that are specific to modeling the immune system itself.

2 The Immune System

The vertebrate immune system (IS) is a multi-layer system that protects the body against disease. The immune system consists of cells, tissues, and organs working together in a complex and highly regulated fashion to repel foreign organisms from the body. A complete synopsis of the immune system is far beyond the scope of this paper, however, it is important to understand the basic parts and functions in order to understand the simulation model.

The immune system actually is composed of two systems, the innate and adaptive systems. The innate immune system consists of physical and chemical barriers at the skin and the mucus membranes at various openings into the body. Combined with secretions such as tears, sweat, and mucus, this part of the immune system seeks to prevent entry into the body. When foreign microbes do enter the body, they are generally eliminated by general purpose cells, called macrophages.[24]

If the defenses presented by the innate immune system are insufficient, a second immune response is initiated. This adaptive response involves the production of specific antibodies against a particular disease causing agent, or pathogen.[24] Production of these antibodies is, for obvious reasons, of great interest to the medical community, understanding and guiding that process is the foundation of immunology. However, because this process also involves communication in a mobile network and knowledge discover in a distributed system, it is of theoretical interest to the computer science community.[63] Accordingly, almost all work in immune system simulation has focuesed on this secondary, adaptive response. The adaptive immune system will be explored in some detail below.

2.1 The Adaptive Immune System

The cells of the immune system originate in the bone marrow. Precursor cells generated in the bone marrow differentiate into all of the cellular elements of blood, including those of the immune system. These precursor cells, as they age, differentiate into specialized cells. The first differentiation is between the myeloid line and the lymphoid line. Myeloid cells differentiate into the cells of the circulatory system (red blood cells) and innate immune systems. Lymphoid cells differentiate into the B and T cells of the adpative immune system, known as lymphocytes.[24] Myeloid cells, however, do play a role in triggering the adaptive immune response.

B lymphocytes

2.2 The Immune System as Computational Model

2.2.1 Information Processing in the Immune System

The immune system is an information processor. T cells and B cells are both generated in the bone marrow, in a process that is believed to be essentially random. T cells divide into two classes, the T-helper cells, which activate other cells, and cytoxic T cells, which kill cells that are infected with viruses. This occurs when the T-cell binds to an antigen that is on the surface of the infected cell. Autoimmune disease occurs when the bodies T cells begin to attack healthy tissue.[24]

In a healthy immune system, T cells differentiate between cells that are healthy and cells that are not. Processing of T cells occurs in an organ called the thymus. Newly generated T cells move through the lymphatic system from the bone marrow to the thymus, where they are tested. T cells that react to the bodies native tissue are destroyed, while T cells that function properly mature and proceed into the body.[46]

In this tiny step of immune system functioning, we can see that there is a lot of information processing going on. First, a random process generates three different types of cells. The B cells may or may not produce antibodies when they encounter the bodies own tissue, but they mature in the bone marrow and move into the lymphatic system anyway. T cells, on the other hand, are randomly generated and filtered, in a second processing step, to extract those cells which process the correct information. The reason this works is because B cells will not activate unless they are in the presence of a T-helper cell, so the B-cell relies on the T-cell to process information for it.

Each phase of adaptive immune system response involves information processing tasks such as string recognition, object deletion, marking cells for garbage collection. The computational power of the immune system has not gone unnoticed by computer science researchers.

2.2.2 Artificial Immune Systems

A full discussion of Artificial Immune System (AIS) research is beyond the scope of this paper. AIS has been used in many areas of computer science research. We will examine two examples of AIS to demonstrate how the structure of the immune system, approached as a metaphor, has extended the power of genetic algorithms.

A mobile ad hoc network is a self-organizing network of communication nodes. One problem that needs to be addressed in such networks is the detection of and removal of misbehaving nodes from the network. Traditional approaches focus on known misbehavior patterns or verification schemes to determine whether a given node should be trusted as part of the network.[39] The problem with such schemes is that the misbehavior has to be imagined ahead of time in order to be detected. An AIS model should detect nodes that evidence unanticipated misbehavior, and create new classifications for faulty nodes.[63]

The deployment designed by Sarfinjanovic and Le Boudec uses an AIS model to detect faulty nodes. The nodes undergo a training phase, where it learns to detect self-nodes and differentiate other-nodes. This creates a set of antibodies specific to the node. Each node makes a local decision, when interacting with a neighbor node, as to whether the neighbor is misbehaving. If the neighbor is classified as misbehaving, the node replicates the software antibodies that let it to so classify the neighbor, generating a resistance to similar misbehavior in the future. Their results were encouraging, in simulation, they were able to detect all misbehaving nodes with a false positive rate of .002.

Another area in which an AIS is useful is in system regulation. A simulation study of a large power system has shown that adaptive immune response is effective in regulating large and dynamic systems. In addition, the authors were able to present a formal mathematical model of immune response, which may transfer over to immune system simulation as well.[72]

2.3 Some Early Work in Immune Simulation

Work was done during the 1970's on the modeling and simulation of biological systems, including the immune system. In 1972, DeLand published a paper on the development of BIOMOD, a biological modeling system developed at the Rand corporation. BIOMOD was probably the first serious attempt at a biological simulation environment for use by non-specialists (that is, by biologists) and had an ambitious mission.[13] DeLand hoped to develop a system that allowed a secondary user to define models, based on simple primitives, that described basic biological systems such as portions of the circulatory system. Although a great deal of concept formalization was accomplished, there was no definitive outcome to this research effort and it is still ongoing, as the complexity of the biological systems continues to unfold.

The GOMPSIM simulator was a specific immune system simulator targeted at cancer research.[19] The authors extended an earlier system, CELLSIM,[14] and used the Gompertz equation to model the growth of tumors. The goal of GOMPSIM was to determine the point at which chemotherapy would be the most useful by modeling the life cycle of cells in the tumor. If the release of a drug could be timed to coincide with the point in the life-cycle of the tumor where the maximum number of tumor cells are at their most vulnerable, the drug would be much more effective. The GOMPSIM system authors hoped to be able to model a tumor with the accuracy needed to determine when that point would be. This research is ongoing.[22]

Continuous simulation was show to be effective an very controlled environment.[37] This simulation of antibody-antigen reactions in a test tube successfully used a differential equation result to approximate the real event. This research, while specific to immune system simulation rather than the general case of modeling biology, has less of a heritage in modern immune simulation than BIOMOD or GOMPSIM. The authors use of continuous simulation techniques rather than discrete event simulation runs strongly against the current theory.[30] Also, test tube reactions are an abstraction themselves. Most interest in this area is not on modeling chemical reactions per se, but on modeling specific behavior of a living system.

Early work in simulating the immune system and other biological processes was limited to a great extent by the power of the systems available. There is a strong desire, however, to make the simulations useful to the biologist rather than the computer scientist, to create a modeling language that would work for this sort of simulation. The following discussion of discrete event simulation continues this trend.

3 Parallel Simulation of Discrete Events and Agents

Discrete event simulation is a method of simulation in which the the state of the system is updated only at discrete times. It is particularly suited to simulations with a large number of stochastic or directed events. Discrete event simulation has been used for decades to model complex processes that do not yield well to differential equations. For approximately as long as there has been discrete event simulation, there have been attempts to unify simulation techniques under a single modeling system.[45] Another problem that has been consistently important for the last thirty years is finding an efficient method for event scheduling, with an O(1) data structure being introduced in 2005.[77, 57, 68]

A detailed discussion of Parallel Discrete Event Simulation (PDES) to 1990 can be found in [17]. A difficulty in PDES is the event horizon. Two events, scheduled to execute on two processors, may or may not be safe to execute concurrently. Whether they are or not depends on whether the output of one event should be the input of the other. There are three strategies for avoiding the causal error of executing an event before its input has been created. The first is to have an event window of 0, where only events that are scheduled to occur at the time of the global clock are executed. This is not an efficient use of PDES, because many events may be scheduled to occur within a very short, but non-zero, period of time. These events would be executed sequentially when they could have been executed in parallel. The second is conservative simulation. A conservative simulation employs a look-ahead scheme that determines what events are safe to execute. The third is optimistic simulation, which employs a rollback scheme to revert the system to a previous state if a causal error is detected, but executes events forward in a fairly wide window otherwise.

3.1 Conservative Simulation of Discrete Events

As mentioned, conservative PDES is a methodology based on prediction. As an example, assume a group of events to simulate, E₁, E₂, E₃ on processors P₁, P₂, P₃, at times t, t+1, t+2. A conservative simulator will determine each events causal chain to determine whether it is safe to simulate that event at some global time. For example, if all three events would finish execution at time t+4, it is safe to execute all three events. If E₁ finishes at t+1, it would be safe to execute E₁, E₂. Different strategies exist to determine how far the simulation looks ahead, if it is determined whether E₃ depends on either E₁ or E₂, and how data is stored and accessed.

This leads to an obvious problem, if an event is mistakenly simulated, the entire simulation will be invalidated. Also, it is possible to create causal deadlocks, where one event depends on another, but neither can be simulated until the other. This event deadlock is a serious problem in PDES.[55] Efforts have been made to extend new data structures for use in both optimistic and conservative event simulation. The Parallel heap is one data structure that offers O(logn), access, this was not improved on substantially until ladder queue, which is O(1).[68] Both data structures create a framework that eliminates deadlock.

Recent work in conservative PDES has focused more on standardization and usability than efficiency, [64] present a simulator which is meant to be used primarily by non-experts in PDES. They focus on the fact that PDES is not broadly accepted in the simulation community. Because of the specific needs of a lookahead algorithm, it is difficult for simulationists to design code that is easily parallelized unless they are already experts in parallel event simulation. The authors hope that user transparent model library will solve this problem. However, they acknowledge that some domain specific models will be needed. The authors vision is that general simulationists will be able to use the general purpose system, requesting domain specific models as needed. This leaves open the question of allowing non-expert users from other areas to design simulations.

The System for Parallel Agent Discrete Event Simulator is a conservative, agent based simulator for implementation on a grid, based on a common message passing standard.([61]) The SPADES architecture depends only on each agent in the simulation being capable of utilizing standard UNIX pipes. Intelligent agents execute on the host system, with explicit modeling of their sensing, thinking, and acting components. This sense-think-act loop is central to the design. Each agent receives its input from a communication server, returns a set of actions to the server, and follows this up with a message that it is done thinking about its current input. This software loop is a crucial component of the lookahead algorithm. SPADES is presented as being mostly useful for artificial intelligence simulation, but there is no reason that it could not be adapted for other work.

Theoretical work in conservative PDES suggests that there are scalability limits.[32] The authors note that PDES has not made a great number of inroads beyond the computer science community, despite the obvious suitability of lattice representation (cellular automata) to many problems in physics and biology and the suitability of such a scheme to conservative event simulation. Because of the large scale of such problems, the authors are interested in what the limits of conservative PDES are as the number of processing elements or events approaches infinity. What they found was that while the simulation was arbitrarily scalable, because some events would be far ahead of other events, extracting information at some particular time in the simulation was not scalable. This implies that a useful simulation, one with good reporting capability, would need to tune the simulations event window, sacrificing scalability for reporting.

3.2 Optimistic Simulation of Discrete Events

In optimistic PDES, the simulation engine does not perform any lookahead services. Instead, events are simulated according to the global clock with no regard for causality, and any causality errors are detected after they occur. If this happens, the affected part of the simulation must rollback to a point before the error occurred, and the events proceed with the causality error adjusted for. Naturally, research in PDES is primarily concerned with two areas: avoiding and executing rollback.[17]

Global event queues allow for greater speedup than local event queues in an optimistic simulator.[57] This is in direct contrast to conservative simulation, where global event queues carry a performance penalty. In optimistic PDES, however, speedup is achieved in part by reducing the rollback occurrence, and a properly designed global event queue does just this. However, reducing rollback is only part of efficiency in optimistic PDES, in addition, there is a space consideration, because the state of each variable must be saved at the point the simulation must roll back to.

One method for simplifying state saving, is by making use of persistent data structures.[6] Because time stamps can be used to determine the version number of a state change, and deprecated states can be discarded efficiently, such data structures have a space saving over traditional methods. There is a small time penalty to this approach but it is constant.

Further speedup and efficiency is demonstrated by the use of the SyncSim algorithm,[56], which reduces speedup as well as limiting the number of saved rollback states to one. SyncSim makes use of the Parallel Heap data structure, which is the only non-constant time part of the algorithm. It is possible that altering the algorithm to take advantage of the more recent LadderQ data structure ([68])would improve performance. A key result of SyncSim is that it is superior to even highly optimized alternatives in an out of the box form.

Another approach to optimistic PDES is reverse computation.[69] In reverse computation, a reverse version of the simulator is compiled along with the forward version. This requires special reversible data structures and algorithms, especially for random number generation. When a causal error is detected, the simulation runs backwards to the point before the error occurs, and then forward again. The reverse computation method may be faster under some circumstances than state saving, and can save space. However, this is not a general result, so many areas of simulation would not benefit from the reverse computation method. For this reason, it does not seem to be a likely candidate for use in a general purpose simulation engine.

3.3 Modern General Purpose or Combined PDES

The defense departments High Level Architecture (HLA) is the dominant architecture in the field,[18]. HLA also encourages the reuse of simulation components, and interaction between different simulations. HLA specifies a federation of different simulations, termed federates. These simulations can be implemented in different languages and run on different platforms, provided that they all can access and communicate with the HLA Runtime Infrastructure, which specifies how communication passes between federates and directs the simulation as a whole. An open problem with HLA is allowing non specialists to use the tools in the system, which is difficult to learn.[64]

HLA is superior in many respects to the Common Object Request Broker (CORBA) and the Java based Remote Method Invocation (RMI), primarily because of its history as a simulation specific platform.[7] Like HLA, CORBA is language independent. CORBA uses an Interface Definition Language, similar to C++, to describe the overall application. Once that is done, interfaces can be written in any CORBA supported language or platform, and these will communicate with an Object Request Broker (ORB). So the various interfaces, or objects, can communicate with each other through the ORB, regardless of what language they are implemented in or what platform they are running on. A failing of CORBA for simulation purposes is that it lacks many of the simulation specific tools that HLA has. It is also a large and complicated system.

RMI is platform independent in the sense that any Java Virtual Machine will work as part of a single application, but not language independent. RMI is more secure than HLA or CORBA, but has the failing that it isn't designed for simulation purposes and requires Java.[7]

The WARP IV simulation kernel is a system meant to replace the HLA architecture, for next generation simulation.[65] This layered distributed architecture contains time management algorithms for sequential processing, rollback and lookahead, and so can be used for sequential, optimistic, or conservative simulation. The Warp IV system uses persistent versions of many data structures found in the C++ Standard Template Library. A single modeling framework allows the non-specialist to write simulations for the system. WarpIV shows good speedup between its sequential and distributed algorithms and solves many problems in PDES.

The �sik architecture is presented as being appropriate as a single modeling platform for parallel, sequential, conservative, or optimistic simulations. If this architecture performs well, it could provide a much needed platform for non PDES specialists to efficiently and easily parallelize their simulations. �sik is presented as a micro kernel for simulation. The micro kernel handles only routing of events, which are seen as autonomous entities. This allows the use of a wide variety of simulation techniques, within a single environment, but not one that is as complete as the WarpIV model.[53]

A different approach to any discussed so far is the Discrete EVent Simulator (DEVS). Although technically DEVS is a discrete optimistic system, it is also considered a mixed discrete and continuous system.[81] DEVS is the result of two decades of consistent research and development in modeling and performance evaluation of a general simulation engine.[1] DEVS is a formal structure that separates modeling from the details of the simulation. Unfortunately, it still requires a very solid understanding of DEVS to implement, but the models once constructed are portable to a variety of different platforms. In addition to creating a formal structure for modeling, DEVS also takes advantage of its model based architecture to save processor space. A DEVS model is viewed as a seven-tuple, M={X,S,Y,d_int, d_ext, l, ta}, where X is a set of external events, S a set of states, and Y a set of output events. A key effect of this formalism is that if the model is in a state where S does not change (for example, because X and Y did not change) than the model simply sleeps until it receives an event X that changes S. This more event based model allows for larger simulations of complex systems.[11]

The DEVS system is highly parallelizable because of its compartmentalized structure. Parallel DEVS systems have been in the literature for at least 10 years. Original models for parallelizing DEVS, the P-DEVS approach, were based in compatibility with the HLA architecture previously mentioned, while recent work has focused on DEVS/CORBA for distributed architectures.[80] This work has the great promise for CORBA of removing one of its perceived disadvantages vs. HLA. The HLA architecture is built specifically for simulation. As such, versus CORBA, HLA is more user friendly, providing more system tools for discrete event simulation. A strong DEVS/CORBA interface would provide the needed system tools. The drawback for the general simulation community is that there is not widespread acceptance of the DEVS paradigm.¹

3.4 Summary

There are significant barriers to entry to the simulator wishing to parallelize code.[6] What is needed is a modular parallelizing engine that allows non-specialists to run accurate simulations. This need has been emphasized for several decades in both sequential and parallel simulation.

While there are general purpose engines available, such as DEVS, �sik, HLA, and CORBA, these engines solve many communication issues, but they do not solve the problem that PDES is a specialists tool. In order to use any parallel simulation at this time, the user has to be aware of event windows, choose whether to use optimistic or conservative schemes, and work accordingly. Until parallel code can be quickly and easily generated, these barriers to PDES use will remain in place.

4 Modeling the Immune System

Modeling of the immune system is a subset of the larger field of biological simulation. As such, it is related to work in artificial intelligence, artificial life, neural networks, and network simulation. The problems and limitations that relate to simulation in general are relevant in immune simulation. In addition, there are some specific challenges facing the biological simulationist.

4.1 Modern Work In Biological Simulation

Biologists need good models of biological systems. Not only is it cheaper to work with a computer model than it is to work with living tissue, it can be more practical to test theories in silicon to conserve the more time intensive and sensitive biological experiments. A truly good simulation would be not just a platform for running tests, but also a legitimate source of new information about its own interaction. Compared against the ideal slick, real time, real space, highly parallelized, compartmentalized and user friendly simulation engine envisioned in the 1970's, todays simulations are still very primitive.

There are two reasons why today's systems have not kept pace with the ideals of yesterday. One is that the more we learn about biology, the more there is to model. In a very real sense, the problem has actually gotten harder. A second is that the necessary programming techniques are still in development. The first problem can be solved by hardware alone, and it is likely that it will be either through quantum or nano computing. The second problem remains outstanding, and there is no guarantee that we will be able to use the computer systems of tomorrow any more effectively than we do the systems of today.

This is evident in the field of neurological simulation. While we are approaching computers that have the raw processing power of a human brain[12], we don't have a behavioral simulator that is as complex or capable as the brain of a rat. Our understanding of the underlying algorithms of living systems must improve if we are to develop computers past their current simple state. Modeling is one way to gain this understanding.

The biological simulationist faces a number of specific challenges.[16]. First, biological systems are hierarchal, involving several layers of complex interaction from basic chemical reactions to patient scale effects. The interaction between objects on a given level can be complex, but so can the interaction between objects on different levels. Second, any biological simulation is a simulation of a poorly understood, adaptive system with numerous levels of feedback from many sources. One solution to this problem is to develop complex, hierarchal, agent based simulations. Further, the level of knowledge embedded in the system should be exhaustive. This analysis also stresses the need for and appropriateness of distributed immune simulation. A great deal of processing power is needed for such simulations, and the agent based nature of the simulation is appropriate for a distributed environment.

A platform that addresses this challenge is the multi-agent system. This distributed framework is used in ([28]) to model biological networks. The agent granularity in this implementation is at the level of a molecular species, with reaction events calling for less or more of each species to model particular reactions. The implementation seems counterintuitive, however, because of the enormous number of molecules involved in even a small simulation, some change in granularity was required. The authors application of this model to specific interactions in cellular networks showed strong correspondence to the physical analysis, suggesting that this is a viable approach to biological simulation in general.

4.2 Immune System Simulations

4.2.1 ImmSim

An immune system simulator that has appeared consistently in the literature for almost a decade is the ImmSim system.[9] ImmSimm was originally written in APL, but has since been ported to C, and a parallel version is available as well. The C-ImmSim code is available under the GPL, which is convenient for researchers and may be partly responsible for the systems success.² ImmSim will be discussed in some detail.

The ImmSim system is designed to simulate interaction between cells inside of a singe lymph node. The system is composed of a number of cellular or molecular entities that can take on a number of states based on each entities repertoire. Examples of cellular entities would be Lymphocyte B cells and Lymphocyte killer T-cells, examples of molecular entities are antigens and danger signals. The repertoire of each entity is based on the length of a bit string, this bit string is mutable in B-cells, for example, giving the modeled B-cells a repertoire of 2²¹. The states that each entity can take include active, infected, resting, and dead. So a particular B-cell in the simulation in an active state, if it encounters an antigen in an active state, will be stimulated by the antigen if the antigen is in this particular B-cells repertoire. This will shift the state of the antigen to resting (it has been bound) and eventually to dead, and the state of the B-cell will alter to stimulated, which will give the B-cell a new set of behaviors to choose from.[30]

These environment is represented by a regular triangular lattice (a hex grid), which is a modified cellular automata. Each node in the grid contains some number of entities. During the first phase of a simulation step, each cell of the grid simulates the interactions between the entities it contains. The interactions are stochastic in nature, there is only a probability that particular interactions will take place. During the next step, entities may propagate to adjoining grid sites. The results are tabulated and then the process repeats until the simulation comes to a close.[4]

The parallel version, ParImm, follows this pattern very closely, simply parallelizing each step and allowing entity propagation between processing elements, along with an initialization step. There are some modifications for optimization. In the original APL simulation, the cells are stored in dynamic arrays, with one array element being reserved for each possible member of a given entities repertoire. This would imply for the C-IMMSIM program that there would need to be a static array of size 2²¹ to store B cells. To avoid this problem, the authors store each entity class in a separate linked list. Each processing element keeps its own set of linked lists.

Another issue is the calculation of the hamming distance between strings, which is a necessary component of antigen/receptor calculation. This calculation can be time consuming. The authors solution was to build a hash table containing the number of ones for each number n, 0 < n � 2²¹, using the bitwise or between compared strings to provide a lookup key. The hamming distance between two strings is the entry in the table at that key.[4]

Initial work in the ImmSim family of programs shows that it ports well between languages, is parallelizable, and the parallel version is scalable. The system also has been shown to mimic the behavior of the immune system. ImmSim3 has been used to test theories of infection and vaccination.[60] Results indicate a preferred period of immunization, based on the response of their simplified immune system. An additional result was the exploration of synergy between innate and adaptive immune response.

A follow up study using the same system, but with different parameters, verified this result. A study of cancer vaccination conducted in 2005 illustrated that there is an effective schedule for vaccination. An unexpected result was the discovery that general, broad spectrum vaccination may actually have a deleterious effect on immune response to a specific vaccine. This dual result, of learning about the system itself, as well as exploring a specific medical hypothesis, is emblematic of what is hoped for in immune system simulation.[8]

PathSim is a simulator that treats the immune system as an Information Rich Virtual Environment. This tool is a parallel system that models the immune system in much higher detail than any other system presented so far. PathSim models several million agent based cell interactions in a 3-D mesh.[54] The technical details of the immune simulation algorithms and models are not available, but the system is build for a high performance environment.³ The authors mention the ImmSim system in a context with suggests that their machine may be an adaption of the Ceiden/Selada model.

4.2.2 Other Immune System Simulators

There are a number of other immune system simulators in the literature. None of them have the following or results of the ImmSim system, and none of them have been parallelized. They all use slightly different approaches to this difficult problem.

LiSEB, an APL based language for emulating living systems, was an early attempt to use modeling to analyze living systems and to account for the autonomous nature of simple living creatures.[5] In LiSEB, each simple entity is actually modeled as a complex system, accounting for the underlying complexity of even single cells. LiSEB would seem to be a strong candidate for parallelization, however. In LiSEB, each object is autonomous and messages are passed through proxies in this way: each object passes messages to a mailbox agent, who passes them to the recipient. In a distributed system, of course, such an interaction scheme is ideal. Further investigation did not show that this work has been followed up on, the last mention of LiSEB is the initial 1994 paper.

The SIMMUNE system presents the immune system as a 3-Dimensional network of cells. This is a fully agent based simulation, with little or no concern for performance but mostly concern for representational accuracy. The cells are not abstracted to finite state machines, they are viewed in terms of stimulus and response. SIMMUNE models the emergent complex behavior of the immune system as part of a larger class of complex systems.[41, 40]

CAFISS, the Complex Adaptive Framework for Immune System Simulation, is a Java based simulator that is designed to be event, rather than centrally, controlled. Using Java's Delegation Event Model, the authors were able to build a simulation system that is modular. Users can easily add or remove cell types from the system without disrupting the core simulator. The goal of the CAFISS simulator is similar to that of the SIMMUNE model, the immune system is seen as a complex adaptive system and modeled within that paradigm. However, the CAFISS team is more concerned with immune research. Their system has successfully been used to replicate results from biology in modeling HIV infection, and the authors note that the modularity of their system should allow for the easy introduction of novel or improved cell types, vaccines, and antigens.[70]

5 Conclusion

The immune system is immensely complex. Though some simulators boast the ability to simulate millions of cells, this claim is relatively modest compared to the billions of cells actually involved in the biological system. It will be a long time before a truly accurate immune simulator can be built to run on any computer. Parallelization is crucial if any progress is going to be made in this area.

For thirty years, there has been one goal of biological modeling: modular, accurate, easy to use systems that can be understood by and used by biologists and immunologists to run in silico experiments. This goal is hampered by two factors. First, the aforementioned complexity of the system makes it difficult to build simple models. Second, simulationists who are not specialists in parallel coding cannot create or work with PDES simulations. If it isn't easy for a programmer, a simulation expert, to create accurate PDES simulations, how can this be simple for the biologist. A slick interface can apply a bandaid to the problem, but in the end the biologist needs to be able to create simulations, not just run preset simulations with variable parameters.

A next generation immune simulator needs to meet the following goals. First, it must accept entities in the form of a common biology language, such as Systems Biology Markup Language. Second, it must be scalable from laptops to supercomputers without user initiated fine tuning. Third, it should be agent based and adaptive in accordance with best previous results and the nature of the system to be simulated.

References

[1]: D.-K. Baik and B. P. Zeigler, ``Performance evaluation of hierarchical distributed simulators,'' in WSC '85: Proceedings of the 17th conference on Winter simulation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1985, pp. 421--427.
[2]: L. P. Baldwin, T. A. Eldabi, and R. J. Paul, ``Simulation modeling as an aid to decision-making in healthcare management: the adjuvant breast cancer (abc) trial,'' in WSC '99: Proceedings of the 31st conference on Winter simulation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1999, pp. 1523--1531.
[3]: R. V. Belani, S. M. Das, and D. Fisher, ``Modeling very large scale systems: one-to-one modeling and simulation of unbounded systems: experiences and lessons,'' in WSC '02: Proceedings of the 34th conference on Winter simulation.1em plus 0.5em minus 0.4emWinter Simulation Conference, 2002, pp. 720--724.
[4]: =2plus 43minus 4M. Bernaschi and F. Castiglione, ``Design and implementation of an immune system simulator,'' Computers in Biology and Medicine, vol. 31, no. 5, pp. 303--331, 2001. [Online]. Available: =0pt
[5]: P. Bottoni, M. Mariotto, and P. Mussio, ``LiSEB: a language for modeling living systems with apl2,'' in APL '94: Proceedings of the international conference on APL : the language and its applications.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1994, pp. 7--16.
[6]: D. Bruce, ``The treatment of state in optimistic systems,'' in PADS '95: Proceedings of the ninth workshop on Parallel and distributed simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 1995, pp. 40--49.
[7]: A. Buss and L. Jackson, ``Distributed simulation modeling: a comparison of HLA, CORBA, and RMI,'' in WSC '98: Proceedings of the 30th conference on Winter simulation.1em plus 0.5em minus 0.4emLos Alamitos, CA, USA: IEEE Computer Society Press, 1998, pp. 819--826.
[8]: F. Castiglione, F. Toschi, M. Bernaschi, S. Succi, R. Benedetti, B. Falini, and A. Liso, ``Computational modeling of the immune response to tumor antigens: implications for vaccination,'' in Decision and Control, 2004. CDC. 43rd IEEE Conference on, 2005.
[9]: F. Celada and P. Seiden, ``Modeling immune cognition,'' in Systems, Man, and Cybernetics, 1998. 1998 IEEE International Conference on, vol. 4, 1998, pp. 3787--3792.
[10]: G. Chen, Z. Li, D. Yuan, and Nimazhashi, ``A model of multi-agent system based on immune evolution,'' in AINA '05: Proceedings of the 19th International Conference on Advanced Information Networking and Applications.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2005, pp. 53--58.
[11]: A. C. H. Chow and B. P. Zeigler, ``Parallel devs: a parallel, hierarchical, modular, modeling formalism,'' in WSC '94: Proceedings of the 26th conference on Winter simulation.1em plus 0.5em minus 0.4emSan Diego, CA, USA: Society for Computer Simulation International, 1994, pp. 716--722.
[12]: D. DeGroot, ``Modeling and simulating the brain as a system,'' in PADS '05: Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2005, p. 3.
[13]: E. C. DeLand, ``Biological primitives,'' in Proceedings of the 1972 SIGGRAPH seminar on Computer graphics in medicine.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1972, pp. 134--137.
[14]: C. E. Donaghey, ``CELLSIM: A digital simulation language for cell kinetic modelling,'' in ACM '78: Proceedings of the 1978 annual conference.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1978, pp. 684--697.
[15]: L. B. Ellwein and D. T. Purtilo, ``Simulation modeling of lymphocyte proliferation and lymphoma development in immune compromised patients,'' in WSC '92: Proceedings of the 24th conference on Winter simulation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1992, pp. 1021--1027.
[16]: P. K. Fink and L. T. Herren, ``Modeling disease processes for drug development: bridging the gap between quantitative and heuristic models,'' in WSC '96: Proceedings of the 28th conference on Winter simulation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1996, pp. 1183--1190.
[17]: R. M. Fujimoto, ``Parallel discrete event simulation,'' Commun. ACM, vol. 33, no. 10, pp. 30--53, 1990.
[18]: ------, ``Parallel simulation: distributed simulation systems,'' in WSC '03: Proceedings of the 35th conference on Winter simulation.1em plus 0.5em minus 0.4emWinter Simulation Conference, 2003, pp. 124--134.
[19]: V. K. Ghanta, C. D. Atnip, M. White, and R. N. Hiramoto, ``A proposed simulation system for modeling cell growth and assessing the effect of drug treatment on tumors,'' in ACM-SE 14: Proceedings of the 14th annual Southeast regional conference.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1976, pp. 202--208.
[20]: N. Giambiasi, J.-L. Paillet, and F. Ch�ne, ``Simulation and verification II: from timed automata to devs models,'' in WSC '03: Proceedings of the 35th conference on Winter simulation.1em plus 0.5em minus 0.4emWinter Simulation Conference, 2003, pp. 923--931.
[21]: J. P. Goldman, W. J. Gullick, D. Bray, and C. G. Johnson, ``Individual-based simulation of the clustering behaviour of epidermal growth factor receptors.'' in SAC '02: Proceedings of the 2002 ACM symposium on Applied computing.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2002, pp. 127--131.
[22]: R. Hu and X. Ruan, ``A simple cellular automaton model for tumor-immunity system,'' in Robotics, Intelligent Systems and Signal Processing, 2003. Proceedings. 2003 IEEE International Conference on, vol. 2, no. 2.1em plus 0.5em minus 0.4emIEEE, 2003, pp. 1031--1035.
[23]: J. Hunt and D. Cooke, ``An adaptive, distributed learning system based on the immune system,'' in Systems, Man and Cybernetics, 1995. 'Intelligent Systems for the 21st Century'., IEEE International Conference on, vol. 3, 1995, pp. 2494--2499.
[24]: C. A. Janeway, P. Travers, M. Walport, and M. J. Shlomchik, Immunobiology: The Immune System in Health and Disease, 5th ed., S. Gibbs, Ed.1em plus 0.5em minus 0.4emNew York, NY, USA: Garland Publishing, 2001.
[25]: N. Kam and D. Harel, ``The immune system as a reactive system: modeling t cell activation with statecharts,'' in Human-Centric Computing Languages and Environments, 2001. Proceedings IEEE Symposia on, 2001, pp. 15--22.
[26]: J. Keane and L. Atlas, ``Molecular signal processing and detection in t-cells,'' in Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on, vol. 2, no. 1597-1600, 2002.
[27]: J. F. Keane, C. Bradley, and C. Ebeling, ``A compiled accelerator for biological cell signaling simulations,'' in FPGA '04: Proceedings of the 2004 ACM/SIGDA 12th international symposium on Field programmable gate arrays.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2004, pp. 233--241.
[28]: S. Khan, R. Makkena, F. McGeary, K. Decker, W. Gillis, and C. Schmidt, ``A multi-agent system for the quantitative simulation of biological networks,'' in AAMAS '03: Proceedings of the second international joint conference on Autonomous agents and multiagent systems.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2003, pp. 385--392.
[29]: T. Kiesling and S. Pohl, ``Time-parallel simulation with approximative state matching,'' in PADS '04: Proceedings of the eighteenth workshop on Parallel and distributed simulation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2004, pp. 195--202.
[30]: S. H. Kleinstein and P. E. Seiden, ``Simulating the immune system,'' IEEE MultiMedia, vol. 2, no. 4, pp. 69--77, 2000.
[31]: S. H. Kleinstein and J. P. Singh, ``Why are there so few key mutant clones? the influence of stochastic selection and blocking on affinity maturation in the germinal center,'' in International Immunology, vol. 15, no. 7, Japanese Society for Immunology.1em plus 0.5em minus 0.4emJapanese Society for Immunology, 2003, pp. 871--884.
[32]: G. Korniss, M. A. Novotny, A. K. Kolakowska, and H. Guclu, ``Statistical properties of the simulated time horizon in conservative parallel discrete-event simulations,'' in SAC '02: Proceedings of the 2002 ACM symposium on Applied computing.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2002, pp. 132--137.
[33]: =2plus 43minus 4D. P. Kotler, ``The Biology of Human Immunodeficiency Virus Infection,'' Nutr Clin Pract, vol. 19, no. 4, pp. 324--329, 2004. [Online]. Available: =0pt
[34]: R. Laubenbacher, ``A computer algebra approach to biological systems,'' in ISSAC '03: Proceedings of the 2003 international symposium on Symbolic and algebraic computation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2003, pp. 5--6.
[35]: Y. J. Lee, S. K. Kim, and K. S. Lee, ``Auto steering control of unmanned container transport (UCT) with vision system and mediated immune algorithm controller,'' in Industrial Electronics Society, 2004. IECON 2004. 30th Annual Conference of IEEE, 2004.
[36]: Z. Li, J. Wu, and Z. Mao, ``Application of artificial immune algorithm in the dynamic zoning of elevator traffic,'' in Intelligent Control and Automation, 2004. WCICA 2004. Fifth World Congress on, 2004.
[37]: B. M. Liu and J. Steensgaard, ``Computer aided analysis of antigen-antibody reactions,'' in ACM-SE 14: Proceedings of the 14th annual Southeast regional conference.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1976, pp. 188--193.
[38]: J. C. Lowery, ``Getting started in simulation in healthcare,'' in WSC '98: Proceedings of the 30th conference on Winter simulation.1em plus 0.5em minus 0.4emLos Alamitos, CA, USA: IEEE Computer Society Press, 1998, pp. 31--36.
[39]: H. Luo, J. Kong, P. Zerfos, S. Lu, and L. Zhang, ``Ursa: ubiquitous and robust access control for mobile ad hoc networks,'' IEEE/ACM Trans. Netw., vol. 12, no. 6, pp. 1049--1063, 2004.
[40]: M. Meier-Schellersheim, ``Understanding information processing in the immune system; computer modeling and simulations,'' in Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on, vol. 4.1em plus 0.5em minus 0.4emIEEE, 2002, pp. 4036--4039.
[41]: =2plus 43minus 4M. Meier-Schellersheim and G. Mack, ``SIMMUNE, a tool for simulating and analyzing immune system behavior,'' Unknown, 1999. [Online]. Available: =0pt
[42]: G. A. Mills-Tettey and L. F. Wilson, ``Security issues in the abels system for linking distributed simulations,'' in ANSS '03: Proceedings of the 36th annual symposium on Simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2003, p. 135.
[43]: C. Monks, P. Crossno, G. Davidson, C. Pavlakos, A. Kupfer, C. Silva, and B. Wylie, ``Three dimensional visualization of proteins in cellular interactions,'' in Visualization '96. Proceedings., 1996, pp. 363--366,503.
[44]: M. Nagasaki, A. Doi, H. Matsuno, and S. Miyano, ``Integrating biopathway databases for large-scale modeling and simulation,'' in CRPIT '04: Proceedings of the second conference on Asia-Pacific bioinformatics.1em plus 0.5em minus 0.4emDarlinghurst, Australia, Australia: Australian Computer Society, Inc., 2004, pp. 43--52.
[45]: R. E. Nance, ``Towards a unifying structure for discrete event simulation,'' SIGSIM Simul. Dig., vol. 3, no. 2, pp. 4--9, 1972.
[46]: NIH, ``Understanding the immune system: How it works,'' Online, 2003.
[47]: S. Oeda, T. Icmmura, and T. Yamashita, ``Biological immune system by evolutionary adaptive learning of neural networks,'' in Evolutionary Computation, 2002. CEC '02. Proceedings of the 2002 Congress on, vol. 2, 2002, pp. 1976--1981.
[48]: M. Oprea and S. Forrest, ``Simulated evolution of antibody gene libraries under pathogen selection,'' in Systems, Man, and Cybernetics, 1998. 1998 IEEE International Conference on, vol. 4, 1998, pp. 3793--3798.
[49]: H. Ortega and M. Martin-Landrove, ``A model for continuously mutant HIV-1,'' in Engineering in Medicine and Biology Society, 2000. Proceedings of the 22nd Annual International Conference of the IEEE, vol. 3, 2000, pp. 1917--1920.
[50]: Y. Osana, T. Fukushima, and H. Amano, ``ReCSiP: a reconfigurable cell simulation platform: accelerating biological applications with fpga,'' in ASP-DAC '04: Proceedings of the 2004 conference on Asia South Pacific design automation.1em plus 0.5em minus 0.4emPiscataway, NJ, USA: IEEE Press, 2004, pp. 731--733.
[51]: M. Osborn and S. Wolpert, ``A vlsi-based biological interface for extracellular potentials,'' in Bioengineering Conference, 1993., Proceedings of the 1993 IEEE Nineteenth Annual Northeast, 1993, pp. 184--186.
[52]: J. H. Pardue, J. Thomas D. Clark, and J. Armand B. Cognetta, ``A systems analysis and model of real-time skin cancer treatment,'' in WSC '95: Proceedings of the 27th conference on Winter simulation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1995, pp. 1054--1059.
[53]: K. S. Perumalla, ``�sik- a micro-kernel for parallel/distributed simulation systems,'' in PADS '05: Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2005, pp. 59--68.
[54]: N. F. Polys, D. A. Bowman, C. North, R. Laubenbacher, and K. Duca, ``Pathsim visualizer: an information-rich virtual environment framework for systems biology,'' in Web3D '04: Proceedings of the ninth international conference on 3D Web technology.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2004, pp. 7--14.
[55]: S. Prasad and N. Deo, ``An efficient and scalable parallel algorithm for discrete-event simulation,'' in WSC '91: Proceedings of the 23rd conference on Winter simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 1991, pp. 652--658.
[56]: S. K. Prasad and Z. Cao, ``Parallel distributed simulation and modeling methods: Syncsim: a synchronous simple optimistic simulation technique based on a global parallel heap event queue,'' in WSC '03: Proceedings of the 35th conference on Winter simulation.1em plus 0.5em minus 0.4emWinter Simulation Conference, 2003, pp. 872--880.
[57]: S. K. Prasad and B. Naqib, ``Effectiveness of global event queues in rollback reduction and load balancing,'' in PADS '95: Proceedings of the ninth workshop on Parallel and distributed simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 1995, pp. 187--190.
[58]: C. Priami, ``Simulation and verification I: design environments for complex systems,'' in WSC '03: Proceedings of the 35th conference on Winter simulation.1em plus 0.5em minus 0.4emWinter Simulation Conference, 2003, pp. 897--905.
[59]: P. Prueitt, ``Continuous analogs to discrete dynamical systems with application to modeling biological response,'' in Neural Networks, 1990., 1990 IJCNN International Joint Conference on, vol. 3.1em plus 0.5em minus 0.4emMadison, WI: International Neural Networks Society, 1990, pp. 529--534.
[60]: =2plus 43minus 4R. Puzone, B. Kohler, P. Seiden, and F. Celada, ``Immsim, a flexible model for in machina experiments on immune system responses,'' Future Generation Computer Systems, vol. 18, no. 7, pp. 961--972, 2002. [Online]. Available: =0pt
[61]: P. F. Riley and G. F. Riley, ``Next generation modeling III - agents: Spades --- a distributed agent simulation environment with software-in-the-loop execution,'' in WSC '03: Proceedings of the 35th conference on Winter simulation.1em plus 0.5em minus 0.4emWinter Simulation Conference, 2003, pp. 817--825.
[62]: A. Rundell, R. DeCarlo, V. Balakrishnan, and H. HogenEsch, ``Systematic method for determining intravenous drug treatment strategies aiding the humoral immune response,'' in Biomedical Engineering, IEEE Transactions on, vol. 45, 1998, pp. 429--439.
[63]: S. Sarafijanovic and J. Y. Le Boudec, ``An artificial immune system approach with secondary response for misbehavior detection in mobile ad hoc networks,'' Neural Networks, IEEE Transactions on, vol. 16, pp. 1076--1087, 2005.
[64]: V. Solcany and J. Safarik, ``The lookahead in a user-transparent conservative parallel simulator,'' in PADS '02: Proceedings of the sixteenth workshop on Parallel and distributed simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2002, pp. 11--16.
[65]: J. S. Steinman, ``The warpIV simulation kernel,'' in PADS '05: Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2005, pp. 161--170.
[66]: C. A. Stewart, ``Introduction,'' Commun. ACM, vol. 47, no. 11, pp. 30--33, 2004.
[67]: B. K. Szymanski, Y. Liu, and R. Gupta, ``Parallel network simulation under distributed genesis,'' in PADS '03: Proceedings of the seventeenth workshop on Parallel and distributed simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2003, p. 61.
[68]: W. T. Tang, R. S. M. Goh, and I. L.-J. Thng, ``Ladder queue: An o(1) priority queue structure for large-scale discrete event simulation,'' ACM Trans. Model. Comput. Simul., vol. 15, no. 3, pp. 175--204, 2005.
[69]: Y. Tang, K. S. Perumalla, R. M. Fujimoto, H. Karimabadi, J. Driscoll, and Y. Omelchenko, ``Optimistic parallel discrete event simulations of physical systems using reverse computation,'' in PADS '05: Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2005, pp. 26--35.
[70]: J. C. Tay and A. Jhavar, ``CAFISS: a complex adaptive framework for immune system simulation,'' in SAC '05: Proceedings of the 2005 ACM symposium on Applied computing.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2005, pp. 158--164.
[71]: N. Toma, S. Endo, K. Yamada, and H. Miyagi, ``An immune optimization inspired by biological immune cell-cooperation for division-and-labor problem,'' in Computational Intelligence and Multimedia Applications, 2001. ICCIMA 2001. Proceedings. Fourth International Conference on, 2001.
[72]: =2plus 43minus 4H. Wang, H. Li, and H. Chen, ``Application of cell immune response modelling to power system voltage control by statcom,'' IEE Proceedings - Generation, Transmission and Distribution, vol. 149, no. 1, pp. 102--107, 2002. [Online]. Available: =0pt
[73]: Y. Wang and C. Wang, ``Modeling the effects of timing parameters on virus propagation,'' in WORM '03: Proceedings of the 2003 ACM workshop on Rapid malcode.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2003, pp. 61--66.
[74]: K. Webb and T. White, ``Combining analysis and synthesis in a model of a biological cell,'' in SAC '04: Proceedings of the 2004 ACM symposium on Applied computing.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2004, pp. 185--190.
[75]: S. Wei, J. Mirkovic, and M. Swany, ``Distributed worm simulation with a realistic internet model,'' in PADS '05: Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2005, pp. 71--79.
[76]: J.-Y. Wu and Y.-K. Chung, ``Artificial immune system for solving generalized geometric problems: a preliminary results,'' in GECCO '05: Proceedings of the 2005 conference on Genetic and evolutionary computation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2005, pp. 329--336.
[77]: F. P. Wyman, ``Improved event-scanning mechanisms for discrete event simulation,'' Commun. ACM, vol. 18, no. 6, pp. 350--353, 1975.
[78]: Y. Xie, Y. M. Teo, W. Cai, and S. J. Turner, ``Servicing provisioning for hla-based distributed simulation on the grid,'' in PADS '05: Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation.1em plus 0.5em minus 0.4emWashington, DC, USA: IEEE Computer Society, 2005, pp. 282--291.
[79]: G. Zaiyi, H. H. Kwang, and T. J. Cing, ``Sufficiency verification of HIV-1 pathogenesis based on multi-agent simulation,'' in GECCO '05: Proceedings of the 2005 conference on Genetic and evolutionary computation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 2005, pp. 305--312.
[80]: B. P. Zeigler and H. S. Sarjoughian, ``Distributed simulation: creating distributed simulation using DEVS M&S environments,'' in WSC '00: Proceedings of the 32nd conference on Winter simulation.1em plus 0.5em minus 0.4emSan Diego, CA, USA: Society for Computer Simulation International, 2000, pp. 158--160.
[81]: B. P. Zeigler and S. Vahie, ``DEVS formalism and methodology: unity of conception/diversity of application,'' in WSC '93: Proceedings of the 25th conference on Winter simulation.1em plus 0.5em minus 0.4emNew York, NY, USA: ACM Press, 1993, pp. 573--579.

1: This perception of DEVS is based on a presentation given by Dr. Ziegler at Georgia Tech in the Spring of 2006
2: the Gnu Public License is one of a large number of open source licenses that allow software developers to release code to the public.
3: This research is being undertaken at Virginia Tech, which has one of the world's top 50 supercomputer installations, composed of 2200 2.3 Ghz 64 bit processors.

This document was translated from L^AT_EX by H^EV^EA.

Contact me with questions:
John P. Daigle