Graph Neural Network Surrogates to leverage Mechanistic Expert Knowledge towards Reliable and Immediate Pandemic Response

Code

• github.com/scicompmod/memilio
  none★ 67

Abstract

During the COVID-19 crisis, mechanistic models have guided evidence-based decision making. However, time-critical decisions in a dynamical environment limit the time available to gather supporting evidence. Infectious disease dynamics are often heterogeneous on a spatial or demographic scale, requiring appropriately resolved models. In addition, with a large number of potential interventions, all scenarios can barely be computed on time, even when using supercomputing facilities. We suggest to couple complex mechanistic models with data-driven surrogate models to allow for on-the-fly model adaptations by public health experts and decision makers. We build upon a spatially and demographically resolved infectious disease metapopulation model and train a graph neural network for data sets representing prevaccination phases of a pandemic. The resulting networks reached an execution time of a fraction of a second, a speeding up the metapopulation up to four orders of magnitude. The approach yields large potential for on-the-fly execution and, thus, facilitates integration into low-barrier web applications for use in pandemic decision-making.

Tasks

Reproductions