Researchers claim new AI tool can help future vaccines "prevent mutations"

The system, known as EVEscape, was developed at Harvard Medical School and the University of Oxford. In testing, the tool accurately predicted the most concerning and common mutations of the SARS-CoV-2 virus that emerged during the pandemic.

A recent study published in the journal Nature revealed a series of promising results. EVEscape's predictions were more accurate than experimental methods and faster and more efficient than lab-based testing. The tool also successfully identified treatment methods for difficult-to-control new variants.

These predictions have had an impact on pandemic surveillance efforts. For over a year, researchers have been publishing biweekly rankings of the risk posed by new SARS-CoV-2 variants. The research findings have been shared with various organizations, including the World Health Organization (WHO).

"There are still thousands of new virus strains appearing every month, too many to test experimentally," said Pascal Notin, a researcher at the University of Oxford and co-author of the study, to TNW. "EVEscape allows us to quickly determine the threat level of new virus strains."

Notin and his colleagues have also successfully predicted mutations of HIV and influenza using EVEscape. They are now testing the tool on lesser-known viruses that have the potential to cause epidemics, such as Nipah and Lassa viruses.

In the future, researchers envision EVEscape guiding vaccine design. Currently, vaccines and treatments are tested retrospectively based on mutations from past epidemics.

EVEscape can provide assessments on the potential directions of viral development, offering hope for creating vaccines that can combat mutations.

How EVEscape Predicts Virus Mutations

This new tool is based on a generative model called EVE (Evolutionary Variation Effect model).

Initially, EVE was used to predict the risk of human diseases, such as cancer, caused by gene mutations. When COVID-19 demonstrated its ability to mutate beyond treatment limitations, researchers adapted their model for SARS-CoV-2.

The generative model has unique advantages in this work. The key to predicting which mutations will evade the immune system is whether they can maintain the so-called "fitness" of viral proteins. Fitness leads to the expression, folding, and binding of functional proteins with host cell receptors.

"Generative models trained on evolutionary sequences have been very helpful in supporting this prediction," said Notin.

To make predictions, EVE is trained to learn concise representations of protein sequences. These sequences have maintained fitness throughout thousands to millions of years of evolution.

"By doing so, it must implicitly learn the biochemical constraints that support functional sequences," explained Notin. "We can then use this understanding to predict new forms of proteins, whether they are human or viral, and whether they themselves are functional."

AI in the Future Battle Against Viruses

EVEscape's adaptability stems from its simplicity. The tool learns from a dataset that only contains viral protein sequences and their structures. Therefore, it can be applied to any virus and used in the early stages of an epidemic.

"This is a huge improvement over the typical approach that requires waiting for relevant epidemic antibodies to be widely available for testing," said Notin.

Another advantage of EVEscape is its modular design. When more powerful generative models emerge, the current fitness prediction model can be quickly replaced.

COVID-19 has also expanded the potential uses of EVEscape. The pandemic has provided valuable lessons for AI prediction and accumulated a wealth of data. With this rich information, EVEscape can make surprisingly accurate predictions for viruses that have been less studied due to a lack of data, which could be highly valuable in future outbreaks.