Speaker
Description
Agent-based models (ABMs) of disease spread are capable of capturing the complexity of host populations, such as behavioural heterogeneity that influences virus dispersal. However, the complex spatial patterns generated by high-dimensional ABM simulations introduce challenges when applying analytical techniques to interpret ABM outputs across parameter space. Here we present a computational pipeline that combines lattice-based ABMs with topological data analysis (TDA) to characterize epidemic dynamics across parameter space without requiring analytical tractability. We implement a stochastic SIR model on a lattice and introduce a locality parameter to continuously interpolate between fully local and fully global transmission. TDA summaries of ABM simulations are computed across a wide range of locality, infection, and recovery rates. Dimensionality reduction is used to embed TDA summaries in $R^3$ and visualize topological differences in parameter space via RGB color mapping. K-means clustering on the resulting embedding further identifies distinct groupings corresponding to qualitatively different epidemic outcomes. Under fully global transmission, the bifurcation boundaries delineated by our TDA approach recapitulate the mean-field prediction that epidemics transition from extinction to endemicity at $R_0 = 1$. As the locality parameter increases, bifurcation boundaries derived from TDA diagrams shift to correspond to a lower effective $R_0$.
