Artificial Intelligence Model May Help Predict Future Viral Outbreaks

Researchers from Harvard Medical School (HMS) and the University of Oxford have developed an artificial intelligence (AI) tool capable of predicting how a virus could evolve to escape the immune system, according to a study published this week in Nature.

The model, known as EVEscape, leverages biological and evolutionary information to forecast future viral mutations and new variants, which the researchers noted could help inform the development of vaccines and therapies for rapidly mutating viruses.

The tool is based on an earlier model designed to explore gene mutations that cause human diseases, called the evolutionary model of variant effect (EVE). EVE uses generative modeling to predict protein functionality based on large-scale evolutionary data across species.

In previous research, EVE successfully enabled the research team to differentiate between benign and disease-causing mutations associated with conditions such as heart disorders and cancer.

“You can use these generative models to learn amazing things from evolutionary information — the data have hidden secrets that you can reveal,” explained senior author Debora Marks, PhD, associate professor of systems biology in the Blavatnik Institute at HMS, in a news release.

During the COVID-19 pandemic, SARS-CoV-2’s ability to rapidly evolve inspired the researchers to expand the EVE model.

“We underestimate the ability of things to mutate when they’re under pressure and have a large population in which to do so,” Marks noted. “Viruses are flexible — it’s almost like they’ve evolved to evolve.”

The research team modified EVE’s generative model—using broader evolutionary information about viruses of interest in combination with biological data about how each virus targets the immune system—to develop EVEscape.

This approach gives EVEscape a more flexible framework, which the researchers indicated can be applied to any virus.

In their most recent study, the researchers assessed whether the model could forecast the most concerning new variants that occurred during the COVID-19 pandemic using only data from before January 2020.

The results demonstrated that had EVEscape been deployed at the beginning of the pandemic, it would have predicted the most frequent mutations and concerning variants of SARS-CoV-2.

The tool’s predictions achieved similar accuracy to that of experimental approaches designed to forecast mutation prevalence based on SARS-CoV-2’s ability to bind to antibodies made by the immune system.

EVEscape’s predictions were also made more quickly than those of lab-based approaches, sifting through thousands of new SARS-CoV-2 variants and identifying those most likely to become problematic. The tool’s efficiency could help inform earlier public health decisions, the researchers noted.

The research team also found that EVEscape could be generalized to other viruses, making accurate predictions about both influenza and human immunodeficiency virus (HIV).

Now, the researchers are applying the model to predict future SARS-CoV-2 variants of concern in real-time. Every two weeks, the team releases a ranking of new variants of concern, which other researchers and public health organizations can leverage. EVEscape’s complete code is also freely available.

Moving forward, the research team aims to utilize the model to investigate less-studied viruses, such as Nipah and Lassa, which both have pandemic potential.

The researchers also seek to use EVEscape to evaluate therapies and vaccines for current and future viral variants.