Bacteriophages are a type of virus that infect bacterial cells. Their ability to do so depends on both the characteristics of their own proteins and those of the potential host. In this master's thesis, we address the problem of embedding the genomes of bacterium-bacteriophage pairs to predict whether a phage can interact with a bacterial cell by entering and manipulating its functions. As the genomes may be of considerably varying lengths, we represent the host-phage pair interaction with the protein pairs that characterise the interaction with their involvement. To accomplish this, we mainly rely on protein language models, which are capable of predicting properties of a given protein from its amino acid sequence. The embedding of pairs obtained with this method may be used to predict whether a species of bacterium may be vulnerable to a specific bacteriophage infection. Our best model achieves an ROC AUC of 0.709. The model would require more training data to increase its reliability.
|
