Inferring contact network characteristics from epidemic data via compact mean-field models

Andrés Guzmán 1 , Federico Malizia 1 , Gyeong Ho Park 2 , Boseung Choi 2,3,4 , Diana Cole 5 , István Z. Kiss 1,6 · 2025

Read Inferring contact network characteristics from epidemic data via compact mean-field models by Andrés Guzmán 1 , Federico Malizia 1 , Gyeong Ho Park 2 , Boseung Choi 2,3,4 , Diana Cole 5 , István Z. Kiss 1,6 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-12040.txt.

Introduction

Andrés Guzmán and co-authors show how to infer contact network traits from epidemic data. Their paper appeared in 2025. Network models beat simple compartment models because they track who meets whom. But the network itself is usually unknown. The team couples a compact network model with a statistical method that fits the whole course of an outbreak. Together they estimate both disease parameters and network features.

Tests on simulated outbreaks recovered the hidden network well, even though structure and disease traits are tangled. The team then fitted two real outbreaks: foot-and-mouth disease in Britain in 2001 and COVID-19 in Seoul. Fits were good, and short-term forecasts held up. The authors say outbreak curves carry fingerprints of the contact web that spread the disease.

Catalog record

CREATORAndrés Guzmán 1 , Federico Malizia 1 , Gyeong Ho Park 2 , Boseung Choi 2,3,4 , Diana Cole 5 , István Z. Kiss 1,6
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE76 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-12040

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Inferring contact network characteristics from epidemic data via compact mean-field models Andrés Guzmán 1 , Federico Malizia 1 , Gyeong Ho Park 2 , Boseung Choi 2,3,4 , Diana Cole 5 , István Z. Kiss 1,6 Abstract Modelling epidemics using contact networks provides a significant improvement over classical compartmental models by explicitly incorporating the network of contacts. However, while network-based models describe disease spread on a given contact structure, their potential for inferring the underlying network from epidemic data remains largely unexplored. In this work, we consider the edge-based compartmental model (EBCM), a compact and analytically tractable framework, and we integrate it within dynamical survival analysis (DSA) to infer key network properties along with parameters of the epidemic itself. Despite correlations between structural and epidemic parameters, our framework demonstrates robustness in accurately inferring contact network properties from synthetic epidemic simulations. Additionally, we apply the framework to real-world outbreaks—the 2001 UK foot-and-mouth disease outbreak and the COVID-19 epidemic in Seoul— to estimate both disease parameters and network characteristics. Our results show that our framework achieves good fits to real-world epidemic data and reliable short-term forecasts. These findings highlight the potential of network-based inference approaches to uncover hidden contact structures, providing insights that can inform the design of targeted interventions and public health strategies. Keywords: Epidemics, Inference, Contact Networks 1 Introduction The spread of infectious diseases is inherently tied to the structure of human interactions. Network theory provides a powerful framework for understanding how diseases propagate by capturing the complex web of contacts between individuals [ 1 , 2 , 3 , 4 , 5 ] . Studies have highlighted how structural properties such as heterogeneity [ 6 , 7 , 8 , 9 , 10 ] , communities [ 11 , 12 , 13 , 14 ] , clustering [ 15 , 16 , 17 , 18 ] , and degree correlations [ 19 , 20 , 21 , 22 ] play a significant role in shaping epidemic dynamics. Epidemic models have traditionally been used to describe and predict disease spread based on assumptions about the underlying contact structure [ 2 , 23 ] . Ultimately, their applicability to real-world processes depends on the availability and quality of data [ 24 ] . These models range in complexity, from classical mass-action approaches, where populations are assumed to mix homogeneously [ 25 ] , to sophisticated network-based frameworks that explicitly incorporate individual-level connectivity patterns [ 26 , 27 , 28 ] . Simpler models offer easier tractability but may overlook key structural features, while more complex models provide richer descriptions but require more detailed input data [ 29 , 30 , 31 , 24 ] . Finding a balance between these aspects is essential for effective epidemic modelling [ 32 ] . A particularly elegant and efficient modelling framework is the edge-based compartmental model (EBCM), which provides a compact yet powerful representation of epidemic processes on networks [ 33 , 34 , 35 ] . Unlike standard compartmental models where incorporating heterogeneity significantly increases model complexity, EBCM encodes network structure and characteristics through probability-generating functions, allowing epidemic dynamics to be described with only a few parameters and a reduced number of equations. In many real-world scenarios, direct measurements of contact networks are unavailable or incomplete. Although collecting data from contact networks is feasible in certain cases, such as sexually transmitted infections [ 36 , 37 , 38 , 39 ] , it remains challenging for respiratory diseases [ 40 , 41 , 42 ] . While epidemic models are often used to simulate outbreaks given a known network structure, inferring the structure of the contact network from observed epidemic data represents an equally important challenge [ 43 , 44 , 45 , 46 , 47 ] . Since spreading dynamics inherently reflect network properties, they can be used to extract valuable information about the underlying structural information. Various methods have been proposed to reconstruct networks from data, including likelihood-based optimization approaches [ 48 , 49 , 50 , 51 , 52 ] and Bayesian inference techniques [ 53 , 54 , 55 , 56 , 57 ] . However, these methods often require detailed temporal data or strong prior assumptions, making them difficult to apply in real-world epidemic surveillance [ 58 ] . An alternative approach, Dynamical Survival Analysis (DSA), has been introduced to estimate epidemic parameters using infection and recovery time distributions [ 59 ] . Originally developed for mass-action models [ 60 , 61 ] , DSA was recently extended to network-based models [ 62 ] , enabling parameter estimation while incorporating some aspects of network structure. However, existing applications remain limited in their ability to fully capture the heterogeneity of contact networks. In this paper, we integrate the DSA approach with the edge-based compartmental model (EBCM) [ 33 ] to develop a Bayesian framework for inferring both disease and network parameters from epidemic data. This extends previous works [ 63 ] , shifting from identifiability analysis to active inference in both synthetic and real-world scenarios. The manuscript is structured as follows: Section 2.1 introduces the EBCM framework, and Section 2.2 details the Bayesian inference procedure. Section 3 presents validation on synthetic and real data, specifically the first wave of COVID-19 in Seoul and the 2001 foot-and-mouth disease epidemic in the UK. Finally, Section 5 discusses the implications of our findings. 2 Methods In this section, we outline the methodologies that form the foundation of our inference framework. First, we introduce the edge-based compartmental model (EBCM), which provides a compact representation of SIR processes on networks. This model serves as the backbone for describing the epidemic dynamics in structured populations. Next, we present the complete inference process, detailing how these methods are integrated to estimate both epidemic and network parameters from observed outbreak data. Specifically, we employ Dynamic Survival Analysis (DSA) to construct the likelihood function, leveraging its ability to handle censored and aggregated epidemic data. Moreover, we describe the Robust Adaptive Metropolis (RAM) algorithm, a Markov Chain Monte Carlo (MCMC) technique designed for efficient exploration of the parameter space. RAM adapts to the local structure of the posterior distribution, improving convergence and robustness in high-dimensional settings. Together, these methods form a comprehensive framework for inferring epidemic dynamics and network structures from real-world outbreak data. 2.1 Edge-based compartmental model We consider a Susceptible-Infected-Recovered (SIR) epidemic process, where individuals can be in one of three states: susceptible ( S ), infected ( I ), or recovered ( R ). Infection occurs at rate \beta along a link between a susceptible and an infected node, while infected nodes recover independently of the network at rate \gamma . In this study, we employ the Edge-Based Compartmental Model (EBCM) [ 33 ] , which provides a compact and analytically tractable representation of epidemic dynamics on contact networks. The EBCM assumes that disease transmission occurs on a network generated by the configuration model (CM) [ 64 , 65 ] , which is characterized by a degree distribution P(k) . The key idea behind EBCM is to track the probability that a randomly chosen node remains susceptible rather than explicitly tracking individual infection events. A central variable in the model is \theta(t) , defined as the

Read the complete text