Composite method for fast computation of individual level spatial epidemic models

2025-09-04

AI Summary

Individual-level spatial epidemic models (ILMs) incur prohibitive computational costs for large populations and high-dimensional covariate settings. To address this, we propose the Composite Individual-Level Model (C-ILM): first, individuals are clustered into low-interference subpopulations using a Dirichlet Process Mixture Model (DPMM), which better captures heterogeneous spatial infection structures than conventional clustering methods (e.g., K-means); second, inter-cluster transmission is explicitly modeled via four “spike functions”, enabling parallel computation and composite likelihood inference. Evaluations on simulated data and the 2001 UK foot-and-mouth disease outbreak demonstrate that C-ILM achieves several-fold speedup over standard ILMs while improving model fit (e.g., WAIC) and posterior predictive accuracy. Thus, C-ILM establishes a scalable, high-fidelity paradigm for large-scale spatial epidemic modeling.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Individual-level models, also known as ILMs, are commonly used in epidemics modelling, as they can flexibly incorporate individual-level covariates that influence susceptibility and transmissibility upon infection. However, inference for ILMs is computationally intensive, especially as the total population size increases and additional covariates are incorporated. We propose a composite method, the composite ILM (C-ILM), that clusters the population into minimally-interfered subpopulations, with between-cluster infections enabled through a ``spark function.'' This approach allows for parallel computation of subsets before aggregation. Focusing on C-ILM, we consider four ``spark functions'', and introduce a Dirichlet process mixture modelling (DPMM) algorithm for clustering. Simulation results indicate that, in addition to faster computation, C-ILM performs well in parameter estimation and posterior predictions. Furthermore, within C-ILM framework, DPMM algorithm demonstrates superior performance compared to the conventional $K$-means algorithm. We apply the methods to data from the 2001 UK foot-and-mouth disease outbreak. The results provide evidence that C-ILM is not only computationally efficient but also achieves a better model fit compared to the basic spatial ILM.

Problem

• Accelerates computation of individual-level epidemic models
• Enables parallel processing via population clustering
• Improves parameter estimation and model fitting efficiency

Innovation

• Composite ILM clusters population for parallel computation
• Uses spark function for between-cluster infection modeling
• Dirichlet process mixture algorithm improves clustering performance