Fighting foot & mouth disease: An algorithm could make all the difference

A team of University of Canterbury scientists have set out to develop algorithms to inform biosecurity efforts to tackle biological threats that could have devastating impacts on the environment - and the primary sector.

Algorithms being developed by the Canterbury team will present possible outbreak scenarios, rather than just one that's the "most statistically likely" and update predictions in real-time based on new information fed into the model.

Aotearoa was served a reminder last week of the ever-present threat of invasive pathogens and pests when it emerged that Te Whānau a Hinerupe, the country's largest and most iconic pohutukawa tree located at Te Araroa on the East Coast, is infected with myrtle rust.

Myrtle rust, a type of fungus, was first detected in New Zealand at a Kerikeri nursery in 2017 and has since spread through the North Island and the top of the South Island. An outbreak of foot and mouth disease affecting cattle would cost the economy billions of dollars. Scenario modeling by the Ministry for Primary Industries has found that a large outbreak scenario would "put the New Zealand economy into recession". Real gross domestic product (GDP) could decline by 5%, exceeding the 2.2% fall the country experienced during the global financial crisis.

"If a highly contagious viral infection such as foot and mouth disease were to arrive here, the cost could be upward of $16 billion,” says Alex Gavryushkin University of Canterbury Associate Professor of Data Science and co-leader of the project.

Photo: Annie Spratt, Unsplash

Feeding in real-time data updates

“One of the main things policymakers and scientists need from algorithms in an outbreak is to provide explainable and actionable uncertainties. The current algorithm system presents policymakers with only one scenario, based on it being statistically the most probable. For example, it might be 60% likely that Farm A directly infected Farm B, and that is the avenue policymakers start to investigate," explains Gavryushkin.

“However, Farm A might have infected Farm C, then Farm C infected Farm B, and that was 30 percent likely, but the point estimate algorithm didn’t present this as a possibility as it wasn’t the most likely scenario. Responding under the assumption that there was no Farm C can have massive consequences. We’re developing new algorithms to present a range of possible scenarios that collectively account for close to a 100% chance.”

The team, including University of Auckland, Massey University, and MPI collaborators, will develop a new type of algorithm, online Bayesian algorithms, to improve outbreak response times.

Current algorithms for tracing the transmission of diseases like foot and mouth aren't suited to handling new data constantly coming.

Says Gavryushkin: “Once we have this efficient infrastructure for biosecurity algorithms in place, we will be in a far better position to prevent problems further down the track by doing the difficult, time-consuming pre-computations early on, including before outbreaks start and in parallel to them.”