Superspreading dynamics in epidemics: a deep learning and Bayesian inference framework for detection and characterization

Superspreading events (SSEs) play a pivotal role in shaping epidemic dynamics, often driving large outbreaks and complicating control efforts. These events occur when a small group of individuals, known as superspreaders, infects disproportionately large numbers of people within a short time frame and confined area. Current approaches to studying SSEs frequently rely on retrospective analyses or detailed individual-level data, which may not always be available, particularly in real time. This limitation underscores the urgent need for new methodologies capable of promptly detecting and characterizing SSEs using minimal data. This research addresses this challenge by developing a novel framework for the real-time detection and characterization of SSEs based solely on incidence time series data. The proposed approach integrates deep learning methods with stochastic epidemiological models, enabling the study of SSEs without the need for detailed individual-level data. Specifically, a one-dimensional convolutional neural network (1D-CNN) is employed to identify the time points at which SSEs occur. These identified time points are then incorporated into a chain-binomial Susceptible-Infected-Recovered (SIR) model with a time-varying transmission rate, enabling the modeling of transmission spikes during detected SSEs. To infer parameters related to the magnitude and duration of these events, Approximate Bayesian Computation (ABC) is applied, efficiently addressing the computational challenges of evaluating the likelihood function posed by dynamic transmission behaviors. The methodologies are tested on both synthetic and real-world datasets, demonstrating high accuracy in identifying SSEs and producing reliable parameter estimates. Robustness analyses confirm the model’s effectiveness under various noisy data conditions, including underreporting and reporting delays, though challenges remain in scenarios of extreme uncertainty. Overall, this framework serves as a valuable tool for the real-time monitoring of SSEs, supporting timely public health interventions and more effective resource allocation during epidemics.

Superspreading events (SSEs) play a pivotal role in shaping epidemic dynamics, often driving large outbreaks and complicating control efforts. These events occur when a small group of individuals, known as superspreaders, infects disproportionately large numbers of people within a short time frame and confined area. Current approaches to studying SSEs frequently rely on retrospective analyses or detailed individual-level data, which may not always be available, particularly in real time. This limitation underscores the urgent need for new methodologies capable of promptly detecting and characterizing SSEs using minimal data. This research addresses this challenge by developing a novel framework for the real-time detection and characterization of SSEs based solely on incidence time series data. The proposed approach integrates deep learning methods with stochastic epidemiological models, enabling the study of SSEs without the need for detailed individual-level data. Specifically, a one-dimensional convolutional neural network (1D-CNN) is employed to identify the time points at which SSEs occur. These identified time points are then incorporated into a chain-binomial Susceptible-Infected-Recovered (SIR) model with a time-varying transmission rate, enabling the modeling of transmission spikes during detected SSEs. To infer parameters related to the magnitude and duration of these events, Approximate Bayesian Computation (ABC) is applied, efficiently addressing the computational challenges of evaluating the likelihood function posed by dynamic transmission behaviors. The methodologies are tested on both synthetic and real-world datasets, demonstrating high accuracy in identifying SSEs and producing reliable parameter estimates. Robustness analyses confirm the model’s effectiveness under various noisy data conditions, including underreporting and reporting delays, though challenges remain in scenarios of extreme uncertainty. Overall, this framework serves as a valuable tool for the real-time monitoring of SSEs, supporting timely public health interventions and more effective resource allocation during epidemics.