Mechanism of action-based classification of antibiotics using high-content bacterial image analysis.

Image-based screening has become a mature field over the past decade, largely due to the detailed information that can be obtained about compound mode of action by considering the phenotypic effects of test compounds on cellular morphology. However, very few examples exist of extensions of this approach to bacterial targets. We now report the first high-throughput, high-content platform for the prediction of antibiotic modes of action using image-based screening. This approach employs a unique feature segmentation and extraction protocol to quantify key size and shape metrics of bacterial cells over a range of compound concentrations, and matches the trajectories of these metrics to those of training set compounds of known molecular target to predict the test compound's mode of action. This approach has been used to successfully predict the modes of action of a panel of known antibiotics, and has been extended to the evaluation of natural products libraries for the de novo prediction of compound function directly from primary screening data.