Volume 14 Supplement 6
Simulating the decentralized processes of the human immune system in a virtual anatomy model
© Sarpe and Jacob; licensee BioMed Central Ltd. 2013
Published: 17 April 2013
Many physiological processes within the human body can be perceived and modeled as large systems of interacting particles or swarming agents. The complex processes of the human immune system prove to be challenging to capture and illustrate without proper reference to the spacial distribution of immune-related organs and systems. Our work focuses on physical aspects of immune system processes, which we implement through swarms of agents. This is our first prototype for integrating different immune processes into one comprehensive virtual physiology simulation.
Using agent-based methodology and a 3-dimensional modeling and visualization environment (LINDSAY Composer), we present an agent-based simulation of the decentralized processes in the human immune system. The agents in our model - such as immune cells, viruses and cytokines - interact through simulated physics in two different, compartmentalized and decentralized 3-dimensional environments namely, (1) within the tissue and (2) inside a lymph node. While the two environments are separated and perform their computations asynchronously, an abstract form of communication is allowed in order to replicate the exchange, transportation and interaction of immune system agents between these sites. The distribution of simulated processes, that can communicate across multiple, local CPUs or through a network of machines, provides a starting point to build decentralized systems that replicate larger-scale processes within the human body, thus creating integrated simulations with other physiological systems, such as the circulatory, endocrine, or nervous system. Ultimately, this system integration across scales is our goal for the LINDSAY Virtual Human project.
Our current immune system simulations extend our previous work on agent-based simulations by introducing advanced visualizations within the context of a virtual human anatomy model. We also demonstrate how to distribute a collection of connected simulations over a network of computers. As a future endeavour, we plan to use parameter tuning techniques on our model to further enhance its biological credibility. We consider these in silico experiments and their associated modeling and optimization techniques as essential components in further enhancing our capabilities of simulating a whole-body, decentralized immune system, to be used both for medical education and research as well as for virtual studies in immunoinformatics.
Recent years have witnessed a growing interest in systems biology [1–7]. Not only are biological systems themselves better understood, but increased computational power, visualization environments and more readily available distributed computing enhance the value of modeling and simulation. In the literature so far, there has been little concern regarding more sophisticated visualizations in scientific modeling. Noteworthy efforts in this direction include Harvard's BioVisions project . We take the viewpoint, that simulations should involve a high degree of visual realism; visualization then becomes a key part of our modelling approaches. We present our latest 3-dimensional simulations and interactive visualizations of the decentralized processes in the human immune system.
Using agent-based approaches in simulations is another aspect to increase realism in computer simulations. Rules or simple programs and attributes for agents can then drive the overall dynamics of a system of interacting entities, which result in emergent observable patterns [9, 10]. An agent-based approach allows simulations to incorporate computational versions of the physical interaction rules that are observed directly in nature. While the agent-based approach does not replace traditional mathematical modeling , it rather acts as a strong complement for better understanding complex biological phenomena. Furthermore, coupling agent-based simulations with advanced graphics visualization and intuitive interaction interfaces can appeal greatly to life scientists, who do not have a programming background or any interest in learning new modeling environments. Allowing such biology experts to appreciate the value of computer simulations is key to the advancement and wider acceptance of systems biology [2, 7, 12]. Finally, making virtual experiments more accessible to biologists, immunologists, and medical researchers will facilitate answers to particularly those research questions not achievable through purely laboratory means.
In this work, we present our latest simulation of the decentralized processes of the human immune system [10, 13]. Our simulation consists of different compartmentalized regions -- simulated as agent environments -- communicating with one another to produce high-level emergent effects such as an organism's immunity to harmful pathogens. Each compartment consists of large numbers of agents, with relatively simple behavioural rules, that act collectively in highly sophisticated networks of interactions. We chose the common Influenza A virus infection as the base for our immune system simulation.
Adaptive immune system
The adaptive immune response results in the elimination of various pathogens such as viruses and other foreign particles. It is also responsible for developing a memory response for future infections with the same antigens. The mechanism through which humans develop immunity to disease-causing pathogens is through the cellular interactions of two different branches of immunity: (1) humoral and (2) cell-mediated. Humoral immunity acts via the production and secretion of antibodies which neutralize harmful antigens. On the other hand, cell-mediated immunity functions via the destruction of infected cells, in order to suppress any further spreading of a virus. Both branches have the ability to create memory cells that can prevent secondary infections with previously encountered viruses .
Lymphocytes have a very important role in the adaptive immune system: T cells are responsible for cell-mediated immunity and B cells are responsible for humoral immunity. T cells and B cells are mainly found throughout the different lymphatic organelles, where they are most likely to encounter antigens. Both types of lymphocytes have antigen-specific receptors on their surface that allows them to detect specific antigen molecules. The binding of an antigen to a naive lymphocyte results in the lymphocyte becoming an effector cell, a cell that is directly involved in the immune response.
The infection is initially detected by professional antigen-presenting cells such as macrophages and dendridic cells that are nearby. Upon detection, the antigen-bearing cells migrate to nearby lymph nodes to trigger the immune response. Within the lymph nodes, B cells and T cells interact with the antigen-presenting cells as well as free antigen molecules. Some activated T cells, namely the cytotoxic T cells, travel to the tissue to eliminate infected cells, while a subpopulation, called helper T cells, remain in the lymph node to promote further proliferation of nearby lymphocytes. B cells that develop into plasma cells produce large amounts of antibodies, which travel via the blood stream to the site of infection in order to neutralize virus particles. Neutralization involves the blocking of surface receptors on a virus responsible for entry into a healthy cell. After the infection is successfully defended, some B and T cells remain in the body as long lived memory cells. Upon secondary exposure to the same virus, the memory cells can quickly divide and mobilize to form a swift response, eliminating the virus before any serious effects are noticed by the host organism [14–17].
Immune response to influenza A infection
Modeling and simulating the human immune system
Natural immune system processes have been inspirational for the modeling of adaptive and learning systems in the field of artificial intelligence [19, 20]. We do not consider these machine learning aspects here. Mathematical models, based on differential equations, have been used to simulate changes of concentrations of immune system agents and large-scale regulatory and defensive actions of the immune system [11, 12, 21]. Lee et al.  provide a comprehensive mathematical model of the specifics of Influenza A Infection. They also divide their system into compartments: the lung and the lymphatic system. Communication between compartments is implicit in their mathematical model as there is no difference in scale or any concrete notion of location and distance. While mathematical models do provide accurate results and some conceptual understanding of the overall dynamics, they lack interactive visualizations and are not easily expandable for larger, more integrative systems.
Agent-based modeling, in the form of cellular automata has been used to simulate the processes of the immune system . This work is extended in  by introducing cell-mediated interactions to a purely humoral immunity model.
Jacob et al.  present an immune system simulation using the concept of virtual swarms, where simple, yet large numbers of particles interact in 3-dimensional scenarios. This work is built upon in , where different compartments across an abstracted human body are introduced. Similar to our work, simple agent rules are used to create complex interaction networks across different compartments connected through a data exchange channel. What we have added in the simulations we discuss in this paper are (1) more advanced visualizations, (2) physics-based models that drive agent interactions, and (3) running our immune system models within a 3-D virtual anatomy model of a human male.
For programming the behaviours and properties of agents we use an object-oriented architecture. In this approach, objects can have internal information (such as states) and are able to send and receive certain messages. Furthermore, the inheritance model is of high importance as it allows common behaviors, such as cell motility, to be implemented at a higher level and shared by the sub-class implementations of the different types of individual cells. Bersini provides some insight of how to use object-oriented programming to build the different agents of the human immune system .
Integrating different models across many levels of scale is beginning to eliminate what most researchers consider the "bottom-up" approach, where systems are built from atomic parts. In the "bottom-up" approach systems become problematic when attempting to model a realistic process, because of the extremely large number of entities that simultaneously interact. As suggested by Noble , building such multi-scale models "middle-out" seems to be a more feasible approach. By creating a connected multi-scale system, changes in a process (such as an infection) can ripple across the different levels of abstraction, to inform the whole system efficiently. The Multiscale Systems Immunology (MSI) project attempts to model the immune response over many different levels of temporal and spatial scales  and has been inspirational for our multi-scale models.
LINDSAY composer modeling environment
LINDSAY Composer (LC)  is a software suite for creating hierarchical 3-D models and visualizations. The modeling environment consists of a collection of component engines for physics, graphics, cameras, user interaction, and agent behaviour rules. Each engine is responsible for controlling the execution of its respective components at each step of the simulation. Component engines can be split into two groups: native engines and external engines. Native engines, such as the physics and graphics engines, are always present within the LC framework. In the model discussed here, the Immune System simulation is built as a plug-in to LINDSAY Composer with its own engine loaded at start-up.
ImmuneSystem_EpithelialCell defines the state and interaction rules of the epithelial cell agent.
LCTransformComponent defines the location, size and orientation of the agent in 3-dimensional space.
LCSphereGhostObject allows the agent to be aware of other physics objects inside the radius of the ghost object, which is used to create a virtual awareness space around an object. This object contains a child--the ImmuneSystem_AttractionField component-- which attracts certain other agents (such as viruses) towards the current epithelial cell.
OEMeshNode gives the agent a graphical representation, which contains a 3-dimensional mesh, textures, colours, shadows and animations.
LCRigidBodySphere contains all of the physics related information such as mass, acceleration and collision shape.
The human immune system encompasses processes across a wide range of scale: starting with the organismal level, one can integrate interaction processes at the organ level (e.g., inside the thymus), at the tissue level (e.g., inside lymph nodes), at the intra-cellular interaction level (e.g., between bacteria and macrophages) and, finally, at the inner-cell processes such as protein synthesis and gene regulation. The LC framework allows the developer to integrate different modeling techniques across multiple scales (Figure 2) into one comprehensive model. In the case of the immune system, the simulations at a higher level of scale would include the bone marrow, thymus, lymphatic system and circulatory system, while the simulation at a lower scale would represent individual lymph nodes, site of infection and the process of inflammation.
Agent-based model of the human immune system
Epithelial cells (∈) are susceptible to be infected by an Influenza A Virus (v) with a given probability p int (∈,v) upon direct mesh-to-mesh collision. After infection, the cell lives for another ∈ incubation = 200 iterations, followed by cell death and subsequent release of ∈ virusesreleased = 5 new virus particles. In addition, the lung tissue contains 12 initial immature dendridic cells residing in the tissue. These cells become activated upon contact with viral particles, and are able to migrate to the lymph nodes to activate the B and T lymphocytes.
Our experiments look at two different compartments of the human immune system that are able to communicate and coordinate asynchronously. The simulation of the lymph node is executed on a different computing node than the tissue simulation. Our developed protocol allows us to build large networks of simulation nodes working together to form a large-scale distributed simulation system. The network setup as well as the communication between the tissue and the lymph node are controlled through distributed objects (from the Cocoa Objective-C Library), which not only work over a network but also between threads and CPUs, whether on the same computer or not. Therefore, the system is not restricted to one simulation per machine, but rather one simulation per computing core over a large network, which maximizes efficiency with respect to message exchange.
The information received by each distributed controller affects the state of the simulation environment or, in this case, the state of the immune system engine of the respective simulation. The agents controlled by the engine are then able to use the new state provided by the engine to inform their decisions, aside from their current physical interactions.
The distributed approach to our system of simulation is highly flexible for extending it over multiple scales. The distributed controllers are not restricted to communicating across simulations of the same type, but can connect data from simulations over multiple scales. For example, one can introduce a differential equation-based model of lymphocyte growth and development in the bone marrow and link it to the agent population in the circulation and in the lymph nodes.
A limitation of our system lies in different time offsets that can arise from a distributed simulation. Presently, at each distributed controller we make no adjustment with regard to time. That is, if a node is slowing down (e.g., due to limited computing resources) the other nodes connected to it still run at their normal rates. The discrepancy is somewhat accounted for in the fact that the slow node does not receive any updates, and therefore does not broadcast any updates outside of its own rate of execution. This stabilizes the system such that the nodes that require information receive the information more slowly, thus eventually matching the same "simulation" speed as the slowest node. It is worth noting that the "simulation" speed is not the frame rate or execution time, but rather how fast certain events happen inside the context of the simulation.
Results and discussion
200 time steps
200 time steps (24h)
150 time steps (18h)
probability of infection
number of viruses released
2500 time steps (10.5d)
3000000 time steps (>40yr)
probability of becoming memory
number of clones generated
probability of releasing antibodies
2500 time steps (10.5d)
3000000 time steps (>40yr)
probability of killing infected cell (CD8)
probability of becoming memory
probability of apoptosis (contact with CD4 T cell)
probability of activation (contact with virus)
500 time steps (2.5d)
probability of neutralization of virus
probability of reproduction
Conclusions and future work
This paper presents our latest work on simulating the decentralized processes of the human immune system within the context of the LINDSAY Virtual Human. The LINDSAY Composer environment provides a wide array of state-of-the-art simulation techniques including advanced graphical visualization, realistic physical interactions and intuitive live-interaction interfaces. One of the strengths of our immune simulation is the visualization aspect, because it provides dynamic illustrations on the essential concepts of the well-orchestrated system of agent interactions that eventually create complex emergent behaviour such as the acquisition of natural immunity to harmful pathogens. We have created a framework for distributing multiple simulations over a network to increase the efficiency and scalability of our system. Our hope is that by introducing more simulations that work together in a decentralized fashion, we can achieve an advanced and well-connected network of physiological simulations that can enhance our fundamental knowledge of the human body.
In order to produce agent interaction dynamics that are close to (or even the same as) those found in natural biological processes, we are currently applying evolutionary optimization techniques to fine tune the multitude of parameters of our models. At this point, we have gathered some preliminary, yet promising results with other agent-based models .
We consider in silico experiments and their associated modeling and optimization techniques as essential components in further enhancing our capabilities of simulating a whole-body, decentralized immune system, to be used both for medical education and research as well as for virtual studies in immunoinformatics.
We would like to extend our appreciation to Timothy Davison, the architect and current developer of the Lindsay Composer simulation environment. We also acknowledge the work of our programmers and students of the LINDSAY Virtual Human project: http://lindsayvirtualhuman.org.
This article has been published as part of BMC Bioinformatics Volume 14 Supplement 6, 2013: Selected articles from the 10th International Conference on Artificial Immune Systems (ICARIS). The full contents of the supplement are available online at http://www.biomedcentral.com/bmcbioinformatics/supplements/14/S6.
- Kriete A, Eils R: Computational systems biology. 2006, Academic PressGoogle Scholar
- Fisher J, Harel D, Henzinger TA: Biology as reactivity. Communications of the ACM. 2011, 54 (10): 72-10.1145/2001269.2001289.View ArticleGoogle Scholar
- Walker DC, Southgate J: The virtual cell-a candidate co-ordinator for 'middle-out' modelling of biological systems. Briefings in Bioinformatics. 2009, 10 (4): 450-461. 10.1093/bib/bbp010.View ArticlePubMedGoogle Scholar
- Mishra B: Algebraic Systems Biology: Theses and Hypotheses. LNCS 4545. Edited by: Anai H, Horimoto K, Kutsia T. 2007, Springer, 1-14.Google Scholar
- Banga JR: Optimization in computational systems biology. BMC Systems Biology. 2008, 2: 47-10.1186/1752-0509-2-47.PubMed CentralView ArticlePubMedGoogle Scholar
- Dhar P, Weiss R: Enabling the new biology of the 21st century. Systems and Synthetic Biology. 2007Google Scholar
- Priami C: Computational Thinking in Biology. Transactions on Computational Systems Biology VIII. 2007View ArticleGoogle Scholar
- Harvard BioVisions 2011. [http://multimedia.mcb.harvard.edu/]
- Wolfram S: A New Kind of Science. 2002, Champaign, IL: Wolfram MediaGoogle Scholar
- Jacob C, Litorco J, Lee L: Immunity Through Swarms: Agent-Based Simulations of the Human Immune System. Artificial Immune Systems, Volume 3239 of Lecture Notes in Computer Science. Edited by: Nicosia G, Cutello V, Bentley PJ, Timmis J. 2004, Springer Berlin/Heidelberg, 400-412.Google Scholar
- Salzberg S, Searls D, Kasif S: Computational Methods in Molecular Biology. New Comprehensive Biochemistry. 1998, 32:Google Scholar
- Perelson AS: Modelling viral and immune system dynamics. Nature Reviews Immunology. 2002, 2: 28-36. 10.1038/nri700.View ArticlePubMedGoogle Scholar
- Jacob C, Steil S, Bergmann K: The Swarming Body: Simulating the Decentralized Defenses of Immunity. Artificial Immune Systems, Volume 4163 of Lecture Notes in Computer Science. Edited by: Bersini H, Carneiro J. 2006, Springer Berlin/Heidelberg, 52-65.Google Scholar
- Janeway CA, Travers P, Walport M, Shlomchik MJ: Immunobiology: The Immune System in Health and Disease. 2005, New York: Garland Science, [http://www.garlandscience.com]6Google Scholar
- Goldsby RA, Kindt TJ, Osborne BA, Kuby J: Immunology. 2003, New York: W. H. Freeman, [http://www.whfreeman.com/immunology5e]Google Scholar
- Playfair J, Chain B: Immunology at a Glance. 2001, London: Blackwell Science, [In my home library (2004)], 7Google Scholar
- Virella G: Medical Immunology. 2007, Informa Healthcare, London, Great BritainGoogle Scholar
- Van-Tam J, Shellwood C: Introduction to Pandemic Influenza. 2009, CABI Publishing, Wallingford, Oxon, Great BritainView ArticleGoogle Scholar
- Farmer DJ, Packard HN: The immune system, adaptation, and machine learning. Physica D. 1986, 22: 187-204. 10.1016/0167-2789(86)90240-X.View ArticleGoogle Scholar
- Bagley JR, Farmer DJ, Kauffman AS, Packard HN, Perelson SA, Stadnyk MI: Modeling adaptive biological systems. BioSystems. 1989, 23: 113-138. 10.1016/0303-2647(89)90016-6.View ArticlePubMedGoogle Scholar
- Callard RE, Yates AJ: Immunology and mathematics: crossing the divide. Immunology. 2005, 115: 21-33. 10.1111/j.1365-2567.2005.02142.x.PubMed CentralView ArticlePubMedGoogle Scholar
- Lee HY, Topham DJ, Park SY, Hollenbaugh J, Treanor J, Mosmann TR, Jin X, Ward BM, Miao H, Holden-Wiltse J: Simulation and Prediction of the Adaptive Immune Response to Influenza A Virus Infection. Journal of Virology. 2009, 83 (14): 7151-7165. 10.1128/JVI.00098-09.PubMed CentralView ArticlePubMedGoogle Scholar
- Celada F, Seiden EP: A computer model of cellular interactions in the immune system. Immunology Today. 1992, 13 (2): 56-62. 10.1016/0167-5699(92)90135-T.View ArticlePubMedGoogle Scholar
- Bezzi M, Celada F, Ruffo S, Seiden EP: The transition between immune and disease states in a cellular automaton model of clonal immune response. Physica A. 1997, 245: 145-163. 10.1016/S0378-4371(97)00290-2.View ArticleGoogle Scholar
- Bersini H: Immune System Modelling: Artificial Immune Systems. In Lecture Notes in Computer Science. 2006, 4163: 150-163. 10.1007/11823940_12.View ArticleGoogle Scholar
- Noble D: The Music of Life. 2006, Oxford University Press, [http://www.oup.com]Google Scholar
- Mitha F, Lucas TA, Feng F, Kepler TB, Chan C, Kepler TB, Chan C: The Multiscale Systems Immunology Project: Software for Cell-Based Immunological Simulation. Source Code for Biology and Medicine. 2008, 3 (6):Google Scholar
- Jacob C, von Mammen S, Davison T, Sarraf-Shirazi A, Sarpe V, Esmaeili A, Philips D, Yazdanbod I, Novakowski S, Steil S, Gingras C, Jamniczky H, Hallgrimsson B, Wright B: LINDSAY Virtual Human: Multi-Scale, Agent-based, and Interactive. Advances in Intelligent Modelling and Simulation: Artificial Intelligenec-based Models and Techniques in Scalable Computing. Edited by: Kolodziej J, Khan S, Burczynski T. 2012, 327-349. in pressView ArticleGoogle Scholar
- Crampin EJ: Computational physiology and the physiome project. Experimental Physiology. 2004, 89: 1-26. 10.1113/expphysiol.2003.026740.View ArticlePubMedGoogle Scholar
- Hunter PJ, Viceconti M: The VPH-Physiome Project: Standards and Tools for Multiscale Modeling in Clinical Applications. IEEE Reviews in Biomedical Engineering. 2009, 2: 40-53.View ArticleGoogle Scholar
- von Mammen S, Davison T, Baghi H, Jacob C: Component-based networking for simulations in medical education. Computers and Communications (ISCC). 2010, 975-979. IEEE Symposium on 2010Google Scholar
- Roit I: Essential Immunology. 1994, Blackwell, Oxford, UKGoogle Scholar
- Jr JCA, Travers P: Immunobiology: The Immune System in Health and Disease. 1996, Garland, New YorkGoogle Scholar
- Sarpe V, Esmaeili A, Yazdanbod I, Kubik T, Richter M, Jacob C: Parametric Evolution of a Bacterial Signalling System Formalized by Membrane Computing. IEEE CEC 2010: Congress on Evolutionary Computation. 2010, Springer, Barcelona, Spain: SpringerGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.