The pharmacodynamics of antibiotics are currently almost exclusively defined at the population level. However, recent studies have highlighted that microbial pathogens diversify into different physiological states within their hosts, and that the action of antibiotics can vary dramatically with the physiological state of single cells. Thus, a comprehensive approach to quantifying pharmacodynamics at the single-cell level, across bacterial strains and growth conditions, will likely have a profound impact on the development of novel antimicrobial therapies.
Recently developed microfluidic setups, when used in combination with time-lapse microscopy, allow long-term monitoring of growth and gene expression in single bacterial cells exposed to precisely controlled environments. However, the throughput of such methods is currently highly constrained by the image analysis, which still requires manual curation.
We here propose to harness recent progress in deep learning image analysis methods to develop a fully automated image analysis tool for time-lapse microfluidic data. As a proof-of-principle, we will use our tool in combination with downstream Bayesian probabilistic methods to infer detailed single-cell pharmacodynamics of several antibiotics from measurements of growth inhibition and killing of individual bacteria exposed to antibiotics with different growth conditions and treatment protocols.