Infections caused by a range of bacteria represent a significant medical need that is not being sufficiently addressed by the pharmaceutical industry. M. tuberculosis, the ESKAPE bacteria, and Select Agent bacteria constitute three classes of microbes that are relevant to global health in large part because of their resistance to available therapeutics. Most new antibacterials are developed by classical discovery methodologies, such as randomly assaying small molecule collections for growth inhibition ofthe appropriate bacterium. We have chosen to look at antibacterial drug discovery differently and sought a novel strategy utilizing Bayesian models to discover and optimize small molecule antibacterials that is more efficient. For example, we viewed the M. tuberculosis data generated from these random "screens" as a computational learning opportunity. We have used computational algorithms to analyze what attributes ofthe molecules tested are consistent with activity and inactivity. Significantly, this approach yielded validated models for M. tuberculosis that have predicted actives with comparatively high rates of success. Thus, we propose two important extensions of this technology: 1) the optimization ofthe three most promising antitubercular actives arising from our models and 2) the creation and validation of this Bayesian methodology to uncover novel actives against each ofthe ESKAPE and Select Agent bacteria, which will be subsequently optimized. These optimization processes will afford molecules with significant potential as novel therapeutics.
The rise of infectious diseases such as those due to Mycobacterium tuberculosis, ESKAPE, and Select Agent bacteria necessitates novel drug treatments. We have developed and validated computational techniques that learn from activity and cytotoxicity data sets to significantly accelerate drug discovery. We seek to employ these techniques to the discovery and optimization of novel antibacterials.