Probabilistic program inference in network-based epidemiological simulations

Accurate epidemiological models require parameter estimates that account for mobility patterns and social network structure. We demonstrate the effectiveness of probabilistic programming for parameter inference in these models. We consider an agent-based simulation that represents mobility networks as degree-corrected stochastic block models, whose parameters we estimate from cell phone co-location data. We then use probabilistic program inference methods to approximate the distribution over disease transmission parameters conditioned on reported cases and deaths. Our experiments demonstrate that the resulting models improve the quality of fit in multiple geographies relative to baselines that do not model network topology.Author summary: The ability to create computer simulations of epidemics is important to be able to predict where and when people will be become infected, identify factors which either contribute to or slow disease spread, and test various interventions without risking real lives. However, the conclusions of experiments performed using these simulations are only meaningful in the real world if we can be sure the simulation accurately models what is happening in the real world. We study methods for fitting parameters, such as infectiousness, to real world data so that the disease simulator correctly represents the actual disease. We achieve this using probabilistic programming methods which automatically adjust the parameters of the simulator until its outputs look realistic. Our method can work on very detailed simulators which model individual people interacting at specific locations in different locales whereas other methods can only fit very simple simulators.