AI Accelerates Finding Bacteriophages

Title: "Prediction of Klebsiella phage-host specificity at the strain level"

• Authors: Dimitri Boeckaerts, Michiel Stock, Celia Ferriol-González, Jesús Oteo-Iglesias, Rafael Sanjuán, Pilar Domingo-Calap, Bernard De Baets, Yves Briers
• Published In: Nature Communications
• Date: 22 May 2024
• Volume: 15
• Article Number: 4355

Antibiotics are losing effectiveness against bacteria, posing a global health crisis. Bacteriophages offer a promising natural alternative, as they can target and kill specific bacteria. Nonetheless, identifying the right phage to combat a particular bacterial infection can be a time-consuming process.

Researchers at Berkeley Lab are addressing this challenge by harnessing the power of artificial intelligence (AI). They have developed a machine learning model called 𝗣𝗵𝗮𝗴𝗲𝗛𝗼𝘀𝘁𝗟𝗲𝗮𝗿𝗻, which can predict which phages are likely to infect Klebsiella bacteria, a common cause of hospital-acquired infections. Dr. Dimitri Boeckaerts, a computational biologist at Berkeley Lab and the lead author of the study, emphasizes the immense potential of phage therapy and the impact of their AI model in expediting the process of finding effective phages.

𝗛𝗼𝘄 𝗣𝗵𝗮𝗴𝗲𝗛𝗼𝘀𝘁𝗟𝗲𝗮𝗿𝗻 𝗪𝗼𝗿𝗸𝘀

PhageHostLearn is trained on a large dataset of known phage-bacteria interactions. It analyzes the proteins on the surface of both the phage and bacteria, focusing on the initial step where the phage attaches to the bacterial host. Through understanding these protein interactions, PhageHostLearn can predict whether a specific phage is likely to infect a particular bacterial strain. The model then identifies and prioritizes phages for further testing in the lab, saving significant time and resources. Dr. Boeckaerts notes that the model has shown a hit rate of around 65% to 84%, signifying a significant improvement over traditional screening methods.

𝗧𝗵𝗲 𝗳𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝘁𝗵𝗶𝘀 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵

While PhageHostLearn is a significant step forward, the researchers acknowledge that further work is needed. The model is currently effective for Klebsiella bacteria, but the development of similar models for other common bacterial pathogens is essential. Additionally, expanding the available data on phage-host interactions is crucial for the continued development and refinement of AI models. Dr. Boeckaerts concludes with optimism, believing that AI has the potential to transform phage therapy. He highlights the importance of combining advanced machine learning with the power of natural viruses in our fight against antibiotic resistance and in revolutionizing the treatment of bacterial infections.