Calibration of Time-Dependent Parameters in Compartmental Epidemiological Models Using Physics-Informed Neural Networks: An Application to the Italian Covid-19 Pandemic
Background & aims of study Compartmental models provide simple and efficient tools to analyze the relevant transmission processes during an outbreak, to produce short-term forecasts or transmission scenarios, and to assess the impact of vaccination campaigns. However, their calibration is not straightforward, since many factors contribute to the rapid change of the transmission dynamics during an epidemic. For example, there might be changes in the individual awareness, the imposition of non-pharmacological interventions and the emergence of new variants. As a consequence, model parameters such as the transmission rate are doomed to change in time, making their assessment more challenging. Here, we propose to use Physics-Informed Neural Networks (PINNs) to track the temporal changes in the model parameters and provide an estimate of the individuals in the compartments. Methods & results PINNs are particular neural networks that recently gained attention in many engineering applications thanks to their ability to consider both the information from data (typically uncertain) and the governing equations of the system. The ability of PINNs to identify unknown model parameters makes them particularly suitable to solve ill-posed inverse problems, such as those arising in the application of epidemiological models. Here, we develop a PINN-based approach to estimate the temporal changes in the state variables and parameters of a synthetic SIR model of the first year of the COVID19 Italian pandemic. Our results show that the PINNs can well retrieve the key changes in the transmissivity when the number of infected individuals is relatively high . When considering only data on the reported infections, the solutions present large fluctuations in periods of low transmission, which are partially reduced when considering also data on the hospitalized individuals. Implications The proposed PINN-based approach is a robust and easy-to-implement tool to monitor the spreading of a disease when dealing with uncertain reported data. It provides estimates of the temporal changes in the model parameters, allowing for more accurate short-term forecasts. Moreover, transfer learning, i.e. storing knowledge by training a neural network on a problem and then applying it to a similar one, can yield cheap and fast generalizations of the model to different epidemics.