Simulacres et Simulation: Using GAMA to debug a real-world genetic-epidemiology outbreak simulator

Understanding the process of disease transmission is essential for preventing and mitigating the spread of infectious disease, yet it is impossible to observe exact transmission events in the real world. There is a need to leverage new technologies to generate realistic datasets to help researchers study behavioral determinants of transmission and to design new and validate existing models of outbreak reconstruction and risk prediction. Motivated by this need, we have developed Operation Outbreak (OO), an app-based outbreak simulation platform for live simulation of an epidemic through a real social network (Colubri A, et al. 2020). OO transmits a virtual pathogen through participating smartphones and generates contact tracing data that reveals behavioral patterns of disease transmission. In this work, we demonstrate how to combine GAMA with OO to use the former as an entirely computational agent-based model (whereas in OO the agents are real individuals interacting in the physical world) that enables us to configure and “debug” OO simulations before carrying out large-scale deployments. Recently, our lab demonstrated the utility of OO-generated data in training an epidemiological model of disease transmission on college campuses (Specht I, et al. 2022). The next step in OO involves developing a genetic model to drive the “intra-host” evolution of synthetic pathogen genome sequences stored on the phones, enabling the generation of grown-truth phylogenetic trees capturing the entirety of host-to-host transmissions as well as the intra-host variability of the pathogen. This highly unique data from OO simulations will be useful to evaluate performance of transmission network reconstruction methods and to learn about the properties of epidemics. Existing tools allow to simulate contact and transmission networks and phylogenetic and sequence evolution, such as the FrAmework for VIral Transmission and Evolution Simulation (FAVITES) (Moshiri N, et al. 2019). However, a key difference in our approach is that the epidemic simulation in OO is driven by the real-life social network of the participants and therefore, by their (unpredictable) behaviors. We are following a multi-step approach focused on SARS-CoV-2, where the first step consists in comparing and tuning viral evolution models in FAVITES, prior to the second step of incorporating those models into GAMA’s COMOKIT (Gaudou B, et al. 2020) to run COVID-19 agent-based simulations in college campus scenarios matching our planned real-life deployments. We will customize COMOKIT to support dynamic genome entities that can be modified throughout the simulation to represent the emergence of novel variants due to random mutation. As the final step, the resulting ground truth models will be used to conduct a live OO simulation of the spread of a virtual virus on an entire college campus in the U.S., for which we will recruit a representative cohort including students, faculty, and staff. This result in contact data coupled with the complete phylogenetic tree of the virtual virus. This will represent the first genetic-epidemiological dataset comprising the full description of a simulated live outbreak, including real-world contact data and synthetic viral genomes.

Project Website: https://operationoutbreak.org/