Assessment of simulation-based inference methods for stochastic compartmental models in epidemiological research

Area of Science:

• Epidemiology
• Computational Biology
• Statistical Modeling

Background:

• Global pandemics necessitate accurate stochastic epidemic models to understand disease transmission dynamics.
• Effective public health strategies rely on timely parameter estimation for rapid forecasting.

Purpose of the Study:

• To compare the performance of pseudo-marginal Particle Markov Chain Monte Carlo (PF) and Conditional Normalizing Flows (CNF) for Bayesian inference in stochastic epidemic models.
• To evaluate these methods on standard compartmental models (SIS, SIR, SEIR) with observation models.

Main Methods:

• Utilized Particle Filter (PF) for unbiased likelihood estimation in pseudo-marginal Particle Markov Chain Monte Carlo.
• Employed Conditional Normalizing Flows (CNF) for flow-based posterior approximation.
• Tested methods on Susceptible-Infected-Susceptible (SIS), Susceptible-Infected-Recovered (SIR), and Susceptible-Exposed-Infected-Recovered (SEIR) models.

Main Results:

• Both PF-based likelihood estimation and CNF-based posterior approximation demonstrated accurate and robust inference capabilities.
• The methods effectively captured stochastic epidemic dynamics, offering predictive power for outbreak control.
• Operational robustness was confirmed in a real-world Ethiopian cohort study with noisy, irregular data.

Conclusions:

• Advanced Bayesian inference methods like PF and CNF are effective for parameter estimation in stochastic epidemic models.
• These approaches enhance prediction capabilities crucial for informing public health interventions.
• Publicly available code and data support the application of these methods in real-world scenarios.