• antibiotic resistance
• artificial intelligence
• genomic language models
• phage engineering
• phage therapy

Antimicrobial resistance is a concerning global health threat, with bacterial resistance to antibiotics leading to severe health consequences and increased mortality. The slow and costly development of new antibiotics has redirected attention towards alternative therapies, such as bacteriophage therapy, a very promising therapeutic approach displaying various advantages over the standalone use of antibiotics. However, bacteriophage therapy also has limitations that make it challenging to quickly find a suitable phage meeting all necessary requirements for treating a particular patient. Phage genome engineering is a great opportunity to overcome these limitations and enhance phage antibacterial properties. However, due also to the large presence of genes of unknown function in phage genomes, it is difficult to predict whether a certain genome modification will preserve phage functionality. This work aims to explore this topic by investigating the possibility of using predictions from artificial intelligence models to guide phage genome engineering and eventually validate models' predictions in the laboratory. To this aim, two genomic language models, megaDNA and HyenaDNA, were trained and tested on their ability to distinguish between essential and non-essential gene knockouts. In the laboratory, an attempt was made to perform two different gene knockouts from a selected phage using the yeast-based phage-engineering platform, with the aim of comparing predictions with experimental results. This dual approach aims to provide a comprehensive framework for effective bacteriophage therapy, addressing both the computational and experimental challenges in phage engineering.