FIGURE 3.

A schematic of how machine learning can be used to discover new anti-biofilm agents. A training set comprising peptides or small molecules with known anti-biofilm activity (validated experimentally) are used to train a machine learning algorithm. Chemical features of these molecules (such as size or lipophilicity) can be generated using chemoinformatics methods such as QSAR. The algorithm then creates a mathematical relationship between a variety of features of each molecule and the anti-biofilm activity of the molecule. A second group of molecules (a validation set) with known anti-biofilm activity is then analyzed using the derived algorithm to ensure that it can accurately classify these molecules as having anti-biofilm activity or not. Finally, molecules for which anti-biofilm activity is not known, are then classified by the algorithm, with the output being potentially novel anti-biofilm agents that must then be validated with in vitro or in vivo models.