Machine Learning Guided Approach to Bacteriocin Engineering

Directed evolution (DE) projects have led to many great advancements in the field of protein engineering. Commonly DE applies oversampling of the amino acid sequence space over specific regions of a protein, or else random mutagenesis of the entire sequence. In both cases, the immense potential for diversity means high throughput screening efforts are intense and often costly endeavours. Machine learning algorithms such as random forest, gradient boosted decent and support vector machine can be employed alongside traditional DE methods, to streamline development projects. These powerful algorithms model relationships between key physiochemical and structural properties and the observed protein function to identify the key design features for protein engineering. A prime application of this technology is the design of novel bacteriocins to combat bacterial antimicrobial resistance (AMR). Bacteriocins, which have bacterial origins, are broad inhibitors of many Gram-positive pathogens, such as methicillin-resistant Staphylococcus aureus (MRSA) and are attractive therapeutic options due to their low cytotoxicity and potential for resistance development. Much is to be learned of their exact structure-activity relationship, however it is known that engineering mutations, truncations and chimeras of two or more bacteriocins, can alter their biological function, such that their activity can be tuned or directed towards particular targets. This DE project aims to employ machine learning as a tool to study the structure-activity relationship of a family of bacteriocins, in efforts to better understand re-design these important biomolecules for use as novel antimicrobial therapies.