Investigating immune system aging: system dynamics and agent-based modeling

Grazziela P. Figueredo, Uwe Aickelin

TL;DR

This paper evaluates two distinct simulation paradigms—System Dynamics (SD) and Agent-Based Simulation (ABS)—to model the aging of the human immune system (immunosenescence). While both methods successfully replicate biological trends in T cell population decay, the study reveals a massive trade-off: SD offers lightning-fast systemic insights, whereas ABS provides granular, individual-level realism at a significant computational cost.

What Problem Does This Paper Address?

As we age, our immune system undergoes "immunosenescence," characterized by a decreased ability to respond to new pathogens. A primary driver of this is the shrinkage (involution) of the thymus, the organ responsible for producing "naive" T cells—cells that haven't yet encountered an infection.

The authors seek to determine which modeling approach is best suited for understanding these complex dynamics. Specifically, they want to know:

  1. How fit is an immune system at a certain age?
  2. Can we accurately simulate the transition from thymic-produced T cells to those maintained through peripheral proliferation (cell division outside the thymus)?
  3. Which simulation method—top-down (SD) or bottom-up (ABS)—is more effective for clinical or research use?

Methods

The researchers implemented a mathematical model originally proposed by Murray et al. (2003) using two different simulation techniques within the AnyLogic environment:

  • System Dynamics (SD): A top-down approach using differential equations to represent "stocks" (cell populations) and "flows" (rates of birth, death, and transition). It treats the immune system as an aggregate whole.
  • Agent-Based Simulation (ABS): A bottom-up approach where every T cell is modeled as an individual "agent" with its own state chart (e.g., naive, active, memory). These agents respond to their environment independently.

The models were tested across three scenarios:

  1. No peripheral proliferation.
  2. Peripheral proliferation with no homeostatic reduction.
  3. Homeostatic alteration of death rates (the most biologically realistic scenario).

Datasets

The models were validated using a dataset of T cell receptor excision circles (TRECs). TRECs are circular DNA fragments found in naive T cells that serve as a biological marker; since they don't replicate when a cell divides, their concentration indicates whether a cell came directly from the thymus or via peripheral proliferation.

  • Source: Data from Murray et al. [9] and Comans-Bitter et al. [3].
  • Scope: TREC levels and total CD4+ T cell percentages from 506 individuals across age ranges from 0 to 54 years.

Key Insights and Ideas

  • The Proliferation Pivot: The models confirm that after age 20, the body shifts from relying on the thymus to relying on peripheral proliferation to maintain T cell counts. Without this proliferation, the immune system would collapse much earlier.
  • Aggregate vs. Individual: SD provides an "idealized" version of the system. ABS, by contrast, introduces "noise" because it simulates individual stochastic behaviors. This noise, while computationally expensive, is actually more representative of real-world biological variability.
  • The Complexity Penalty: While ABS allows for modeling interactions between specific cells (heterogeneity), it does not inherently produce more "accurate" results than SD when the underlying mathematical rules are the same. In fact, the SD model had a lower Sum of Square Errors (SSE) in this study because it lacked the "instabilities" found in the agent-based approach.

Key Results

The most striking findings were the differences in computational performance:

  • Processing Speed: SD models were nearly instantaneous, taking roughly 0.7 to 0.8 seconds to simulate 100 years of aging. ABS models took between 8 and 18 hours (28,800 to 64,800 seconds) for the same simulation.
  • Memory Usage: SD required minimal RAM (~11 MB), while ABS required significantly more (~50 MB), scaling with the number of agents.
  • Accuracy: Both methods followed the TREC data trends well in Scenario 3, but SD was more precise in matching the mathematical curve due to its aggregate nature.

Limitations and Future Work

The primary limitation of the Agent-Based Model is its extreme computational demand, making it difficult to use for large-scale populations or real-time clinical decision-making without significant optimization.

Future Directions: * Hybrid Modeling: The authors suggest combining the two—using SD for high-level populations and ABS for specific areas of interest (like a localized infection site) to balance speed and detail. * Patient-Specific Profiles: The ultimate goal is to use these models to define an "Immune Risk Phenotype" for individuals, helping doctors predict how an elderly patient might respond to a specific vaccine or therapy.

Summary updated

Abstract

System dynamics and agent based simulation models can both be used to model and understand interactions of entities within a population. Our modeling work presented here is concerned with understanding the suitability of the different types of simulation for the immune system aging problems and comparing their results. We are trying to answer questions such as: How fit is the immune system given a certain age? Would an immune boost be of therapeutic value, e. g. to improve the effectiveness of a simultaneous vaccination? Understanding the processes of immune system aging and degradation may also help in development of therapies that reverse some of the damages caused thus improving life expectancy. Therefore as a first step our research focuses on T cells; major contributors to immune system functionality. One of the main factors influencing immune system aging is the output rate of naive T cells. Of further interest is the number and phenotypical variety of these cells in an individual, which will be the case study focused on in this paper.
BibTeX
@inproceedings{figueredo2010investigating,
  author = {Grazziela P. Figueredo and Uwe Aickelin},
  title = {{Investigating immune system aging: system dynamics and agent-based modeling}},
  booktitle = {SummerSim},
  year = {2010},
  url = {https://openreview.net/forum?id=TE2T44OlMx},
}

Loading related content...

Figure