Agent-based models (ABMs) simulate biological systems by representing individual entities -- cells, organisms, or molecules -- as autonomous agents that follow local rules, interact with neighbors, and make stochastic decisions based on their internal state and local environment. Unlike equation-based approaches (ODEs, PDEs) that describe populations as continuous variables, ABMs capture the heterogeneity, spatial structure, and discrete nature of real biological systems. Emergent behaviors -- tumor morphology, tissue patterning, collective cell migration, biofilm architecture -- arise from the aggregate of individual agent interactions without being explicitly programmed. ABMs are essential when spatial organization, cell-to-cell variability, and local interactions dominate system behavior.
Biological systems are fundamentally composed of discrete, heterogeneous individuals -- cells with different gene expression profiles, organisms with different genotypes, molecules with different binding states. While ordinary differential equations and partial differential equations have been enormously successful in modeling biological dynamics, they do so by averaging over populations, treating concentrations and densities as continuous variables. This averaging erases exactly the features that drive many important biological phenomena: spatial structure (a cell's behavior depends on which neighbors surround it), stochastic variability (genetically identical cells make different fate decisions), and individual history (a cell that has previously been exposed to a signal behaves differently from one that has not). Agent-based modeling recovers these features by simulating each entity individually.
In a typical biological ABM, each agent (usually a cell) has an internal state (cell cycle phase, gene expression levels, damage accumulation) and occupies a position in 2D or 3D space. At each time step, agents sense their local environment (nutrient levels, signaling molecules, mechanical forces from neighbors), update their internal state according to defined rules (deterministic or probabilistic), and execute actions (divide, migrate, differentiate, apoptose, secrete signals). The rules are local -- an agent interacts with its immediate neighbors and the concentrations at its position, not with the global system state. This locality is biologically realistic: cells do not have access to system-wide information. The population-level behavior -- tissue morphology, growth curves, invasion patterns -- emerges from the aggregate of individual interactions.
ABMs have been applied to a remarkable range of biological problems. In tumor biology, ABMs model the growth of spatially structured tumors with hypoxic cores, proliferative rims, and clonal heterogeneity driven by stochastic mutations -- predicting how tumor architecture shapes drug penetration and resistance evolution. In immunology, ABMs simulate T cell search dynamics in lymph nodes, where the probability of finding a rare antigen-presenting cell depends critically on spatial organization and migration patterns that ODE models cannot capture. In developmental biology, ABMs model morphogenesis -- how sheets of cells fold, branch, and self-organize into organs through local signaling and mechanical interactions. In ecology and epidemiology, individual-based models track pathogen transmission through contact networks, capturing superspreading events and spatial clustering that mean-field models miss.
The main limitation of ABMs is computational cost: simulating millions of individual agents over biologically relevant timescales (days to weeks of real time) requires substantial computing resources, and the stochastic nature of the models demands many replicate simulations for statistical validity. Hybrid models address this by combining ABMs for the components where individuality matters (cells) with continuum equations for the components where it does not (diffusible molecules). For instance, a model of wound healing might use an ABM for fibroblasts and immune cells (where individual migration paths and activation states matter) coupled to reaction-diffusion PDEs for growth factors and oxygen (where molecular individuality is irrelevant). This hybrid strategy balances biological fidelity with computational tractability and represents the current state of the art in multicellular systems biology. Frameworks like PhysiCell, Chaste, and CompuCell3D provide standardized, community-developed platforms for building and sharing biological ABMs.
No topics depend on this one yet.