PROJECT SUMMARY The development of narrow-spectrum antibiotics against the Lyme disease-causing bacterium, Borrelia burgdorferi has the potential to significantly alter current approaches to the treatment and prevention of Lyme disease. Narrow-spectrum agents that affect only the target bacteria avoid issues of propagation of resistance in off-target bacteria, alterations in microbiome, and overgrowth of pathogenic bacteria. However, physical screening of compounds for activity against multiple bacteria such as through traditional high-throughput screens is highly inefficient due to its very low “hit” rate for activity against B. burgdorferi (<0.3%). Even when “hits” are found downstream testing for toxicity, studies of pharmacokinetics, and mechanisms of action are time- consuming and costly. Advances in machine learning can accelerate narrow-spectrum antibiotic development by more efficiently and comprehensively searching the vast potential space of small molecule candidates to identify the optimal balance of inhibitory properties, bio-availability, and toxicity. In this project, we propose to develop two modeling-based platforms to accelerate drug development efforts for Lyme disease using combinations of computational and experimental approaches. The first platform will use machine learning to design compounds with predicted activity against B. burgdorferi but not other bacteria using high throughput screening data from B. burgdorferi, E. coli and S. aureus. This framework will be biologically and chemically informed in an automated way using medical literature agents, predicted proteome binding, and machine learning models of bioavailability and toxicity. The second platform will focus on speeding identification of mechanisms of action of novel agents using a multi-omic profiling model across dimensions of morphology and transcriptional response to known agents to generate predictive models. In the process of developing these tools, our work will also produce d