Scalable contact tracing model with Imperial College to limit COVID-19 spread
With the advent of the pandemic, the need for effective contact tracing models has risen drastically. Creating simulations to realistically represent a city or a country and tracing the interactions between individuals in that network is crucial to understand how the disease behaviour can be controlled and tracked. Countries around the world are trying to implement such models to inform decision-making, such as the NHS Test and Trace system here in the UK.
However, layering vast amounts of data to inform such models is computationally demanding, and with pandemic diseases such as COVID-19, policymakers are interested in national – or even international – scale models, requiring computing power at massive scale. Tracking COVID-19 is even more challenging because the pathogen can start spreading when infected people are still asymptomatic, making it extremely complex to trace the contacts of these individuals. The creation of realistic models is slowed down by the infrastructural limitations of simulations today. As a simulation scales the ineffective provisioning of computational resources for unpredictable occurrences such as convergence zones, large entity counts and data spikes can cause simulations to crash. These are distributed computing problems that burden scientists and get in the way of editing parameters and algorithms in an easy and iterative manner to rapidly obtain impactful insights.
Hadean is working with Imperial College London to produce a scalable contact tracing model to prevent the further spread of COVID-19 in the UK. Using Aether Engine, our spatial simulation library for agent-based modeling, we are able to dynamically allocate compute resources to enable such data-intensive simulations to perform at scale. By taking advantage of its octree data structure to map and allocate virtual space to CPU space, Aether Engine breaks up the workload of complex problems such as an infection network simulation, providing computational power as requested using massively distributed computing across a large number of cloud servers.
The project involves the creation of realistic contact networks using data from the Office for National Statistics where people in a community are represented by the nodes of the network and the relationships between these people by the edges (or links) of the network. Different types of networks can be generated with the model; for example, each person has a social, family, work, and random network which takes into account the disease transmission between socially unrelated agents based on only spatial proximity. By applying this contact network model, the transmission of COVID-19 has been simulated within the UK population, with the ultimate aim of scaling the model to a global level.
Looking forward, it can also be used to test various scenarios such as the effects of international travel or variants on the network, enabling policymakers to evaluate the consequences of actions, before implementing them in the real world. Aether Engine’s cloud-native and scalable nature enables the possibility of rapidly running such scenarios in time-critical situations to inform governments with a complete picture for decision-making.