We have seen firsthand how catastrophic these epidemics can be thanks to the global COVID-19 pandemic, though it could have been far worse. Scientists have now created an AI program that will alert us to potentially harmful mutations in upcoming pandemics.
The early warning anomaly detection (EWAD) system proved successful in forecasting which new variants of concern (VOCs) would arise as the virus evolved when tested against actual data from the transmission of SARS-CoV-2.
EWAD was created by US researchers from Scripps Research and Northwestern University using a machine learning technique. In machine learning, enormous volumes of training data are examined by computers to find patterns, create algorithms, and then forecast how those patterns may behave in upcoming, unknowable scenarios.
In this instance, the genetic sequences of SARS-CoV-2 variations, their frequency, and the reported global fatality rate from COVID-19 were supplied to the AI as infections progressed. The software might then detect genetic changes brought on by the virus’s adaptation, which are often manifested in rising infection rates and declining fatality rates.
William Balch, a microbiologist at Scripps Research, claims that “we could see key gene variants appearing and becoming more prevalent, as well as the mortality rate changing, and all this was happening weeks before the VOCs containing these variants were officially designated by the WHO.”
The team’s particular method, known as Gaussian process-based spatial covariance, essentially runs calculations on a set of historical data in order to forecast future data, taking into account both the averages of the individual data points and the connections among them.
The researchers could demonstrate EWAD’s ability in predicting how interventions like immunizations and mask use could lead a virus to continue evolving by testing their model on something that had previously occurred and discovering close matches between the real and anticipated data.
One of the key takeaways from this research, according to Balch, is the necessity of accounting for tens of thousands of additional undesignated variants, or “variant dark matter,” in addition to a few famous variants.
In order to tackle upcoming pandemics as they arise, the researchers say their AI algorithms were able to identify “rules” of virus evolution that would have otherwise gone undetected.
Additionally, the technology created here might help researchers learn more about the fundamentals of virus life. This might then be applied to enhance medical procedures and other public health initiatives.
Mathematician Ben Calverley of Scripps Research thinks that there are numerous potential future applications for this system and its underlying technical principles.
The research has been published in Cell Patterns.