Predicting phage-host interactions with a layered machine learning approach

The aim of this project, is to develop a prediction model that forecasts which phages can infect which bacterial strains in the context of infections with Antimicrobial Resistant bacterial strains based on their genomic content. The model will be used to better understand phage-bacteria interactions as well as develop phage cocktails for phage therapy. Current statistical models, have identified multi-gene systems that are correlated with phage susceptibility. These multi-gene systems include CRISPR-CAS systems in the bacterial genomes, structural proteins in the phage genomes, bacterial membrane proteins and immunity proteins. These identified systems reflect the infection process of the phage from binding to receptors to infection, replication, and spread. The model that will be developed during the project will include these steps in the process. Reliable and in patient input data is of critical importance for the success of our algorithm. This is needed for both the host and phage data. Secondly, of importance is to efficiently and effectively identify gene systems, for this, new algorithms will be developed during the project. Thirdly, comparative genomics tools such as simultaneous pangenome representations and population structure are vital. Existing tools will be further developed during the project.