Aim 2

Train and test the in silico antibiotic design predictions using experimental screens of bacterial activity

High throughput screening strategies used to test compound libraries for antibiotic activity are often limited to simple, endpoint readouts of e.g. cell growth. This tradeoff improves the speed of screening at the cost of detailed information about a compound’s overall biological activity, missing potential avenues for antibiotic drug development.

In this project, we will take advantage of the latest innovations in robotics-assisted high-throughput microscopy to conduct cell-based “phenomics” screens, to generate experimental data about the bioactivity of different compounds in bacteria. Specifically, our phenomics screens will use combinations of reporter dyes to perform “bacterial cell painting” to visualize changes in key cellular features such as DNA, metabolism, membrane integrity, enzymatic activity, growth, etc., followed over time during drug treatment.

For compounds that produce antibiotic effects (i.e. hits), the phenomics approach provides a cellular “autopsy” that may reveal how the compound functions, by discriminating its effects on distinct features of the bacterial cell. This may help identify the compound’s bacterial target, mechanism of action, its specificity, potential for side effects, toxicity, etc.

Using existing compound libraries, we will first use this phenomics approach to train our active-learning, generative drug design models with information-rich datasets about the biological activity of tens of thousands of different compounds in bacteria. Subsequently, once the models begin making their own viable antibiotic design predictions, this approach will be used to test the predictions and iteratively optimize the models using experimental results.

Image: Bacterial cell painting for “phenomic” screening of bacterial responses to chemical and genetic perturbations. This approach, combined with robotics-assisted high throughput imaging allows the screening of tens of thousands of drug treatments or gene knockouts per day.