Network epidemiology for signal extraction

Traditional epidemic models attempt to fit the spread of infectious disease across populations. However, populations are ultimately an abstraction. At a more fundamental level diseases spread between individuals, not populations. This reflection motivates the network approach. Network approaches in epidemiology help to sharpen our intuitions. For example, they have shown that, in principle, heterogeneity can sustain a disease that would on average go extinct, or asymmetric transmission can alter the effectiveness of interventions. Unfortunately, applying network epidemic modelling directly to data is a challenge since we don't know the relevant contact network, or even how to measure it.   This project will develop computational and mathematical methods for fitting networked epidemic models to empirical epidemiologic data in the regime of high uncertainty. Even when data are of low quality, it may be possible to extract meaningful insights. We may never know exactly who was in contact with whom, or who was infected first, but we may still be able to determine the mode of transmission. And, in the end, this latter question is the scientific question of interest. Algorithms will be derived and validated to lay foundations for data driven approaches.  This project has potential applications for disease control and optimization of interventions. We will focus on whether contact traces – noisy and incomplete at the best of times – contain sufficient signal for answering basic research questions, so that future interventions are optimized. If they do, how do we extract this signal? If not, what modifications would make contact traces more informative without harming their immediate goals?