(346ac) Machine Learning and Computational Simulation Aided Design of Antibacterial Oligothioetheramides

Oligothioethermamides are novel molecules developed to combat the rising crisis of antibiotic resistance. Their antibacterial properties present similarly as antibacterial peptides, and thus they have similar factors that affect their activity and toxicity, namely the backbone hydrophobicity and cationic charge. Our atomistic simulations investigating into two isomeric oligothioetheramides, with the same cationic charge and hydrophobicity have shown that the binding of these oligomers to the surface of lipid bilayers could be a suitable candidate for an objective function that could then be used to aide in their design. Oligothioethermaides have the added advantage of allowing greater control over the backbone sequence, and thus pose a large combinatorial problem that involves exploring the sequence space occupied by the backbone sequence of these oligomers. We have performed coarse-grained simulations in tandem with applying Bayesian optimization to optimize the activity of these molecules and provide suitable candidates for experimentalists. We have also used Support Vector Machines to train on data available in databases for antibacterial peptides. By developing similar data for the combinatorial space of oligothioethermides by mapping and extracting similar parameters, we applied the trained machine on this unexplored space to categorize the oligomers. We thus compared the candidates provided by these two machine learning techniques to measure the performance of our objective function.