Combining an old treatment that has long been abandoned and the promises of artificial intelligence (AI) to combat bacterial infections resistant to antibiotics: this is the challenge of scientists from the Pasteur Institute, Inserm, AP-HP and from Paris Cité University, who have developed a new tool capable of choosing, in a simple and effective manner, the best possible cocktail of bacteriophages for a given patient.
Invented in the 1920s, phage therapy, that is to say the use of viruses called bacteriophages which only infect bacteria to eliminate them in a targeted manner, was abandoned at the end of the 1930s, with the arrival of antibiotics, much easier to use. The approach has been making a comeback over the past twenty years, as antibiotic resistance has grown. However, the great diversity and specificity of bacteriophages make this path complex. “Today, only a few countries in Eastern Europe, such as Georgia, still use phage therapy, while in Western countries, “broad-spectrum” phages are used occasionally on a compassionate basis (as part of of a temporary authorization for nominative use, Editor’s note) to treat chronic infections that are multi-resistant to antibiotics, when no authorized medication is any longer effective.recalls Dr Baptiste Gaborieau, co-first author of the article, resuscitation doctor at Louis-Mourier hospital (AP-HP) and researcher in the IAME laboratory (Paris Cité-Inserm University), in a press release.
Highlighting the crucial role of receptors on the surface of bacteria
In their work published on October 31, 2024 in the journal Nature Microbiology, French researchers provide proof of concept for a tool that would tailor-made bacteriophages based solely on the genome of the targeted bacteria.
First, they first analyzed the mechanisms of interaction between bacteria and phages. The objective: to know if it is possible to predict the effectiveness of a bacteriophage on a given bacterial strain. To do this, they created a database with 403 strains of bacteria on one side. Escherichia coli and on the other 96 bacteriophages. A job that required more than two years of effort. “We brought the phages into contact with the bacteria in culture and observed which bacteria were killed. We studied 350,000 interactions and succeeded in identifying, at the bacterial genome level, the characteristics likely to predict the effectiveness of phages. summarizes Aude Bernheim, main author and head of the Molecular Diversity of Microbes laboratory at the Institut Pasteur.
“Contrary to what was initially thought, it is the receptors on the surface of bacteria and not their defense mechanisms which primarily determine the capacity of bacteriophages to be able to infect bacteria or not, and which predict their effectiveness” , continues Florian Tesson, co-first author of the article and doctoral student (Pasteur and University of Paris Cité-Inserm).
A selection of bacteriophages that is 90% effective
Then the team’s bioinformaticians designed an artificial intelligence program based on the analysis of the bacteria’s genome, and more particularly the regions involved in the coding of the bacteria’s membrane receptors, the entry point for phages. “We are not faced with a ‘black box’ here, and this is what makes the strength of our AI-based model. We know exactly how it works, which helps us improve its performance”underlines Hugo Vaysset, co-first author of the article and doctoral student (Pasteur). After more than two years of training, the AI was able to correctly predict the effectiveness of bacteriophages against bacteria E. coli from the database in 85% of cases, simply by analyzing the DNA of the bacteria. “A result that exceeds our expectations”, recognizes Aude Bernheim.
The researchers tested their model on a new collection of bacterial strains fromE. coli responsible for pneumonia and selected, for each of them, a tailor-made “cocktail” of three bacteriophages. In 90% of cases, the bacteriophages tailor-made by the AI destroyed the bacteria present. “We still need to test how the phages behave in different environments, but the proof of concept is done. We hope to be able to extend it to other pathogenic bacteria, because our AI has been designed to easily adapt to other scenarios, and offer personalized phage therapy treatments in the future. concludes Aude Bernheim.
Related News :