Learning Vaccine Allocation from Simulations

Gerrit Großmann, Michael Backenköhler, Jonas Klesen, Verena Wolf

Abstract: We address the problem of reducing the spread of an epidemic over a contact network by vaccinating a limited number of nodes that represent individuals or agents. We propose a Sim-ulation-based vaccine allocation method (Simba), a combination of (i) numerous repetitions of an efficient Monte-Carlo simulation, (ii) a PageRank-type influence analysis on an empirical transmission graph which is learned from the simulations, and (iii) discrete stochastic optimization. Our method scales very well with the size of the network and is suitable for networks with millions of nodes. Moreover, in contrast to most approaches that are model-agnostic approaches and solely perform graph-analysis on the contact graph, the stochastic simulations explicitly take the exact diffusion dynamics of the epidemic into account. Thereby, we make our vaccination strategy sensitive to the specific clinical and transmission parameters of the epidemic.