Predicting the evolution of viruses with AI: science in pursuit of the invisible

© Anna Shvets/Pexels

In the future, pathogens will cause more regular and more virulent pandemics and several factors argue in this direction. Environmental changes (deforestation, urbanization, climate change, etc.), increased international travel and the phenomenon of antibiotic resistance.

However, this statement should be qualified; advances in epidemiology, molecular biology and public health allow for faster detection of epidemics, a better understanding of transmission mechanisms and the development of new prevention and treatment tools. In this range of tools, AI is one of choice, particularly for deciphering the genetic language of viruses and predicting their future mutations.

Algorithms tested for viral mutations

Let's take the example of RNA viruses; SARS-CoV-2, the archetype of this family, is distinguished by its ability to accumulate genetic modifications over the course of its replications. Unlike DNA, which has correction mechanisms, viral RNA tolerates more errors during its copying, consequently generating a large number of variants.

To better imagine what this means, we can imagine this type of virus as a bouncing ball. With each bounce (infection), it can change a little, making it harder to catch in mid-air. This genetic plasticity allows viruses to explore a vast landscape of evolutionary possibilities. It adapts very quickly to new environmental conditions, such as the emergence of antibodies or antiviral treatments.

To understand this complex dynamic, the research teams at Stanford University, led by Brian Hie and his colleagues, decided to seize the power of LLMs(Large Language Model). Initially designed for textual analysis; like ChatGPT or Claude 3; they adapted them so that they could study viral sequences. These algorithms, fed by millions of genomic data, decipher evolutionary patterns like so many sentences in a molecular language.

This approach not only makes it possible to identify isolated mutations that confer a selective advantage to the virus, but also to anticipate the dominance of certain variants in the viral ecosystem.

200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000

AI, our shield against future pandemics ?

The advent of tools such as AlphaFold from DeepMind (Google) or ESM-2 (Meta) has breathed new dynamism into this field of research, as David Robertson of the University of Glasgow points out. These systems rely on a colossal amount of data – nearly 17 million sequences for SARS-CoV-2 alone. At the University of Tokyo, Jumpei Ito's team developed CoVFit, a predictive model that has already proven itself in anticipating the emergence of certain variants of the virus.

At Harvard University, Debora Marks' team designed EVEscape, a system capable of generating 83 potential versions of the SARS-CoV-2 spike protein. The latter plays a fundamental role in the infection process and has therefore become a primary target for vaccines and treatments. EVEscape can thus create molecular “avatars”, which can be used to assess the effectiveness of future vaccines against possible mutations of the virus.

This tool, like COVfit, has also proven its effectiveness. In March 2024, when the JN.1 variant dominated the global viral landscape, the CoVFit model identified three amino acid changes likely to increase its ability to spread. These mutations were actually observed later, in variants that were gaining ground.

To refine the accuracy of these models, the researchers estimate that at least five years of data on viral evolution will be needed. Teams like Shusuke Kawakubo's in Tokyo are extending their research to other pathogens, including the influenza virus, to anticipate the necessary adaptations of seasonal vaccines.

But (there is necessarily a but) one obstacle remains: that of evolutionary leaps, or the sudden appearance of numerous mutations which give the virus new characteristics. For example: greater transmissibility, an ability to evade antibodies, thus transforming it into a new, potentially more dangerous variant. This was the case with the Omicron variant, which appeared with more than 50 simultaneous mutations; a phenomenon that remains difficult to predict. This is why Robertson and his team are exploring the limits of these unpredictable evolutionary trajectories; to better understand the extent of the possibilities of evolution as soon as a new virus appears.

• RNA viruses evolve rapidly by accumulating mutations, making their monitoring and control more complex
• Artificial intelligence helps anticipate viral mutations and develop tools such as vaccines or adapted treatments.
• Despite its advances, AI still struggles to predict sudden and massive mutations, such as those observed with the Omicron variant.

📍 To not miss any Presse-citron news, follow us on Google News and WhatsApp.

k325.0 reviews

[ ]