Stochastic Individual-Based Simulation of Epidemic Dynamics

32 concepts31 papers1 proposal

Traditional epidemiological models view populations through the lens of aggregate compartments. In a classic Susceptible-Exposed-Infected-Recovered (SEIR) differential equation model, a population is treated as a well-mixed fluid. The model assumes that any given susceptible individual has an equal probability of interacting with any infected individual.

While computationally elegant, this "well-mixed" assumption breaks down when applied to real-world human behavior. Humans do not mix uniformly; we move through highly structured, overlapping social networks shaped by our households, schools, workplaces, and daily commutes. As noted by Eubank et al., aggregate models inherently smooth over the high-resolution heterogeneity that drives real-world outbreaks.

To capture how diseases actually spread, researchers have shifted toward stochastic individual-based modeling (also known as agent-based modeling). Instead of tracking the fraction of a population in a given health state, these simulations instantiate every person as a discrete, autonomous agent. By simulating individual interactions over realistic contact networks, these models allow complex, non-linear population-level phenomena to emerge naturally from simple, local rules.

Networked Probabilistic Finite State Machines

In an individual-based simulation, an agent's health status is not a continuous variable but a discrete state within a network. To operationalize this, frameworks like Simdemics, introduced by Barrett et al., represent individuals as networked probabilistic timed finite state machines.

[ Susceptible ] 
       │
       ▼ (Transition probability determined by contact duration & neighbor attributes)
  [ Exposed ]
       │
       ▼ (Deterministic or stochastic incubation period)
 [ Infectious ]
       │
       ▼ 
 [ Recovered ]

Instead of a deterministic flow rate between compartments, transitions between states are fundamentally probabilistic and local. For example, a susceptible agent does not catch a disease from the "population"; they catch it because they share an edge in a contact network with an infectious agent. The probability of that state transition occurring in a given time step depends on specific, granular variables: the duration of the contact, the transmissibility of the pathogen, and individual attributes of both the host and the source.

Because these individual transitions are stochastic, a single simulation run represents just one possible timeline. By executing numerous parallel simulations, researchers can account for random variability in transmission and clinical outcomes. As demonstrated by Gountas et al., this stochasticity allows the model to generate credible intervals for epidemiological trajectories, capturing the inherent uncertainty of real-world outbreaks.

Structuring the Virtual World with Multi-Layered Networks

An agent-based model is only as good as the environment in which its agents move. To avoid falling back on uniform mixing, researchers construct high-fidelity synthetic populations and contact networks derived from census, demographic, and travel data (Nsoesie et al., Du et al.).

As outlined in the structural taxonomy proposed by Mac Namee et al., these simulation frameworks can be cleanly decomposed into four modular components:

  1. Disease transmission dynamics (the biological rules of the pathogen)
  2. Social population structure (the demographic attributes of the agents)
  3. Agent movement patterns (daily schedules and mobility)
  4. Spatial environment (the physical locations where interactions occur)

By separating these components, researchers can build multi-layered contact networks where agents interact across distinct social spheres, such as families, schools, workplaces, and random community encounters (Romero‐Brufau et al.).

  [ Household Network ]       [ Workplace Network ]       [ Commuting Network ]
    (High density,              (Medium density,            (Dynamic, transient
     long duration)              structured shifts)          random encounters)
          │                             │                            │
          └─────────────────────────────┼────────────────────────────┘
                                        ▼
                           [ Individual Agent Path ]

This structural granularity allows the simulation to capture how localized contact dynamics influence regional risk. For instance, Mac Namee et al. utilized this network architecture to demonstrate that a town's commuting centrality makes it a highly effective regional "super-spreader." Similarly, Adiga et al. integrated standard SEIR frameworks into individual-level simulations of Delhi slums, proving that high population densities and large household sizes fundamentally accelerate epidemics, rendering standard aggregate "stay-at-home" assumptions highly inaccurate.

Why It Works: Capturing Non-Linearity and Interventions

The primary advantage of individual-based modeling is its ability to simulate targeted, non-pharmaceutical interventions (NPIs) and behavioral shifts that are impossible to represent cleanly in differential equations.

Dissecting Emergent Phenomena

Consider the phenomenon of superspreading. In an aggregate model, superspreading is often simplified as a higher average transmission rate. However, Kuylen et al. used an individual-based model to show that the source of superspreading matters immensely. When superspreading is driven by biological infectiousness, outbreaks progress slowly and are easier to contain. Conversely, when driven by high-contact social behavior, it creates explosive epidemic waves with a severe risk of resurgence after lockdowns lift.

Testing Granular Policies

Because agents possess explicit schedules and locations, policy makers can test highly specific, localized interventions:

  • Hybrid Telecommuting: Cohen-Addad et al. simulated disease spread across real-world school and workplace networks to evaluate hybrid schedules. They discovered that "rotating" strategies (splitting populations into alternating weekly shifts) significantly outperform "on-off" strategies by breaking transmission chains before individuals become infectious.
  • Targeted Mask Allocation: Tian et al. modeled individuals as nodes in a contact network to evaluate mask-wearing strategies. They found that "outward-good" masks (which prevent source expiration) are vital early in an epidemic, while "inward-good" masks are crucial later on, and that allocating masks based on an individual's network degree optimizes population-level mitigation.
  • Logistical Optimization: Romero‐Brufau et al. simulated autonomous individuals across multi-layered networks to show that delaying the second dose of an mRNA vaccine for individuals under 65 maximizes population-level survival, provided the first dose maintains high efficacy.

Multi-Scale and Data-Driven Extensions

Modern individual-based simulations frequently bridge the gap between macroscopic population spread and microscopic physical or biological processes.

On a spatial and physical level, Vecherin et al. paired agent-based daily routines with probabilistic microexposure formulas. By simulating specific behavioral archetypes moving through actual workplace floorplans, their model bridges the gap between broad epidemiological trends and high-fidelity aerosol physics.

On a biological level, Durso-Cain et al. coupled stochastic agent-based simulations of population dynamics with internal, deterministic ordinary differential equations (ODEs) that govern intracellular viral replication. This multi-scale approach revealed a powerful synergy: cell-free diffusion seeds new infection clusters, which cell-to-cell transmission then rapidly expands.

[ Macroscopic Scale ]  Stochastic Agent-Based Population Simulation
                             │
                             ▼ (Determines exposure events)
[ Microscopic Scale ]  Deterministic ODEs (Governing intracellular viral replication)

Finally, these models can ingest high-resolution, real-world data to eliminate guesswork. Dong et al. deployed a Bayesian multi-agent framework that combined daily symptom surveys with high-resolution Bluetooth proximity data from smartphones. By tracking actual physical proximity, the model could accurately infer exact transmission chains and predict illness in individuals who failed to report symptoms.

By expanding boundaries beyond isolated environments, these models can also capture complex macro-interfaces, such as tracking how pathogens move between the general community, long-term care facilities, and acute care hospitals (Lin et al.). Ultimately, individual-based simulations succeed because they mirror the real world: they treat an epidemic not as a homogenous population state, but as the emergent consequence of millions of individual, connected choices.

Key Foundational Work

Modeling Infection with Multi-agent Dynamics (2012) — contributed 1 concept elements to this direction.
An Interaction-Based Approach to Computational Epidemiology (2008) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_agent_based_modeling_individual_based_si_83e3b285,
  title = {agent-based modeling (individual-based simulation)},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/agent-based-modeling-individual-based-simulation/rdr_83e3b28554884326a4ebc2fd0b2160ba},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_agent_based_modeling_individual_based_si_83e3b285_v1,
  title = {agent-based modeling (individual-based simulation)},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/agent-based-modeling-individual-based-simulation/rdr_83e3b28554884326a4ebc2fd0b2160ba/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake