Global Health Press

Machine versus virus: Deploying artificial intelligence against future pandemics

How do you design a vaccine against a virus not yet known to man? Doing so could help arm us against future pandemics, by rapidly compressing the time it takes to develop protective vaccines.

Most vaccines work by showing the immune system pieces of viral protein, which are often unstable or fall apart when produced in a lab or manufacturing facility. Designing vaccines against new pathogens requires not only the ability to predict what sort of proteins they might contain, but finding ways to stabilise them.

AI programmes are being used to design prototype vaccines that could rapidly be married to an existing vaccine platform, such as mRNA, to start production of vaccines for clinical testing within weeks of the detection of a new pandemic threat.

Proteins are chains of amino acids that can twist and loop into a bewildering array of structures, and these structures determine how each protein functions, and how our cells – including our immune cells – interact with them.

In recent years, scientists have been developing artificial intelligence programmes that can predict proteins’ 3D shapes based on their amino acid sequences – a fiendishly tricky task that takes weeks or even years to unpick by human hand in the laboratory.

Now AI programmes are being used to design prototype vaccines that could rapidly be married to an existing vaccine platform, such as mRNA, to start production of vaccines for clinical testing within weeks of the detection of a new pandemic threat.

Dr Clara Schoeder was working as a postdoctoral researcher at Vanderbilt University in Nashville, US, when the COVID-19 pandemic hit. The scientists on the benches next to her would go on to develop a cocktail of monoclonal antibodies known as Evusheld, which became an important weapon against SARS-CoV-2. It was through them that Schoeder says she became interested in designing vaccines against viruses with pandemic potential. “It was fascinating to learn what’s out there, and what we need to be prepared for,” she says.

Among those supervising her work was Prof Jens Meiler, co-creator of a biomolecular modelling and design software package called Rosetta, which can quickly and accurately decipher the 3D structure of proteins and aid the design of new ones. “We are using Rosetta to design compounds that don’t exist in nature,” Meiler says.

This includes designing structural features into proteins that can “lock” them into a position that optimises our immune response to them. A similar approach provided the foundation for many of the COVID-19 vaccines in use today, as well as several respiratory syncytial virus (RSV) vaccine candidates, one of which has recently been approved for use in humans – the world’s first.

“In the past, people have tried mutating the amino acids in proteins one by one to try to stabilise them to make better vaccines through trial and error, but that is very hard and time consuming,” says Schoeder.

“The benefit of using a computer is that we can basically sample all of the possible mutations and ask the computer to predict which one of these is likely to be the best one. It could save a lot of time.”

The computer does this by learning from databases of known protein sequences and structures and using these to predict the impact of a change in the sequence of the amino acids on a protein’s structure.

The Institute of Drug Discovery at the University of Leipzig in Germany, where Meiler and Schoeder now work, was recently awarded US$1.9 million by the Coalition for Epidemic Preparedness Innovations (CEPI) to expand application of this software towards the development of vaccines against future pandemic viruses.

To achieve this, they will design protein structures that could be used to build vaccines against existing paramyxoviruses and arenaviruses, including Nipah virus and Lassa virus. “The idea is to identify common concepts shared between proteins from related viruses, which are unlikely to change much between other members of that virus family – including members we don’t know about yet,” says Schoeder, who is leading the project.

Having computationally predicted these protein structures, they will then use AI and Rosetta to devise ways of locking them into conformations that are predicted to trigger an effective immune response.
Some of these designs will then be advanced through pre-clinical and clinical testing, to gather data on their safety and the type of immune response they trigger.

Paramyxoviruses and arenaviruses are two of ten virus families considered to have epidemic or pandemic potential. If a new virus did emerge from one of these families, these protein designs could be rapidly adapted to create vaccine candidates, which would then undergo further human trials.

Source: Gavi – VaccinesWork

Notify of
Inline Feedbacks
View all comments