Solving bacterial infections: How Phagos is using generative AI to change the way infectious diseases are treated

Bacterial infections are the second leading cause of death worldwide. Compounding that, bacteria is growing resistant to antibiotics at an accelerated rate, meaning one in three antibiotics are no longer effective in animal farming.

Antibiotic resistance is a threat for both humans and animals—as there are ten times more animals than humans on earth, most antibiotics are consumed by animals.

"Due to antimicrobial resistance, bacterial infection is expected to be the first cause of mortality by 2050," James says. "There are already millions of human deaths today and billions of animal deaths due to bacterial infection."

Traditional solutions have limitations; new antibiotics can take 10 years and cost several billion dollars to develop. Phagos sought a radically different approach—one that could match the evolutionary speed of bacteria while remaining economically viable.

Phagos is deploying phages, the natural predators of bacteria. Phages are actually viruses that infect and kill only bacteria, not human cells, making them the perfect weapon against antibiotic-resistant infections.

"Bacteria naturally evolve and will become resistant to any stable, fixed substance. But what we do at Phagos is personalize bacteriophage-based treatments that will particularly target a given bacterium to eradicate it,” says James.

However, the technical challenge of this process is immense. The reason phage therapy has not been brought to scale in the past is that matching the right phage to the target bacteria has been largely trial and error.

"Finding the perfect set of bacteriophages to match a given bacterium is like finding a handful of needles in a haystack the size of the galaxy," James explains. “There are trillions of bacteriophages for every grain of sand on the planet."

James explains that you have to screen every single phage against every single bacterial strain, and then, every phage combination against every target bacterial strain combination. “It’s an exponential problem that’s unable to be solved manually.”

Phagos needed a unique AI model that could predict phage-bacteria interactions. Bacteriophages are by nature very specific, meaning they can only eliminate a very small subset of bacteria. The manual process of matching phages to bacteria and then testing them is untenable due to the astronomical number of phages available. AI is the key to automating and accelerating this critical step in developing phage-based therapies.

"We were looking for a cloud provider to train our models, but also for data storage," says James. "We came across AWS and thanks to their broad offer and also the support they were giving us, it was a no-brainer to choose AWS."

Phagos data scientists and bioinformaticians rely on Amazon SageMaker AI to train and fine-tune gen AI models on a significant amount of genomic data, both from public databases and its own lab-generated data.

Phagos uses these AI models in live production. It receives infectious strains in its lab, sequences the bacteria, and then inputs it into the AI engine. The AWS-trained models simulate millions of phage-bacteria interactions and determine the optimal phage characteristics for killing that particular bacteria. Thanks to these AI predictions, Phagos needs 50% fewer tests in the wet lab. It also means a 99.5% time savings when screening phage candidates—10 minutes per bacteria versus 29 hours in the wet lab.

“The sheer scale of what we're processing is overwhelming: massive genomic datasets, AI models trained on biological interactions that no one has ever digitized before, and data from field deployments,” explains James.

“The complexity of working with DNA data also meant that when it came to scaling laws and hyperparameter optimization strategies, we had to start basically from scratch instead of using well known and community-approved recipes,” says James. “The flexibility of AWS allowed us to seamlessly integrate fully managed services with low-level ones, allowing our scientists to build specialized solutions for our specific needs.”

Besides using generative AI on AWS, Phagos’ platform also relies on AWS for infrastructure.

Phagos uses high-performance GPU instances on Amazon Elastic Compute Cloud (Amazon EC2) for the enormous computational power needed to run biological simulations. “AWS is the computational backbone of our entire discovery process; our speed is impossible without it,” says James.

Phagos’ core IP—its growing library of phage and bacteria data—is built on Amazon Simple Storage Service (Amazon S3) as its central data lake. Phagos relies on AWS storage services, specifically Amazon S3 and Relational Database Service (RDS), which give it the flexibility, scalability, and security needed for its comprehensive data platform. While S3 handles the ingestion and storage of large volumes of both unstructured and structured data, RDS manages and facilitates queries of structured data.

Leveraging these services, Phagos has built sophisticated data pipelines that process and store its data in a clean, secure, flexible, and cost-efficient manner. The scalability of S3 and RDS allows Phagos to expand its operations without significant infrastructure modifications. Furthermore, these services facilitate seamless integration with third-party solutions, such as a laboratory information management system (LIMS) and edge/external data providers and consumers.

Phagos has developed its entire data platform on these AWS services, working closely with AWS and AWS consulting partners to optimize its implementation.

James and her Phagos co-founder Alexandros Pantalis began by testing their phage approach in the field. Their destination was an oyster farm in France's Loire Atlantic region where the breeder’s stock was being decimated by a mysterious bacterial infection. None of the antibiotics on hand had worked.

“France is the world’s second-largest oyster producer after China, so economically it was a big deal,” recalls Pantalis. The duo collected samples from the water and the oysters, brought them back to the lab, and identified a phage capable of targeting the culprit bacteria. When they applied it, oyster mortality dropped by 40%. 

Phagos then began building its proprietary AI platform, Alphagos. This year, Phagos received approval from the French Agency for Veterinary Medicinal Products, which means it can now market its solution in France as personalized veterinary medicines, a world first that officially validates the company’s platform and marks a turning point in the fight against bacterial infections. Phagos can regularly update its treatments without having to re-validate the new phage formulation each time.

Phagos has achieved remarkable results since deploying its AI-driven platform and entering its commercial phase.

By using AWS's computational infrastructure, Phagos drastically reduced treatment development time. "Using bacteriophages and AI makes us able to develop a new cure in two months from scratch," says James, “This is a 100 times improvement over traditional antibiotic development.”

The company has grown from two founders to approximately 50 employees. With help from AWS, it has built the infrastructure, vision, and technical ability to develop a model that doesn’t exist anywhere else in microbiology.

Phagos is currently focused on animal farming operations, but they are aiming at human applications. "Addressing animal farming is the first step, but what's nice is that it allows us to generate data and refine our models before we one day go into human phage therapy," James says. "We are currently working on bacteria that are very similar to the ones infecting humans, and we expect to address human phage therapy by 2030."