Physical Biology and Deep Learning for Antibiotic Resistance and Discovery

PROJECT SUMMARY/ABSTRACT The dissemination of antibiotic resistance and the drying-up of antibiotic discovery pipelines threaten to increase morbidity from routine medical procedures and worsen the spread of infectious diseases. The main objective of this project is to address the dual challenges of antibiotic resistance and discovery by (1) pinpointing the cellular pathways involved in antibiotic-induced bacterial cell death and (2) leveraging these mechanistic data to discover and develop novel structural classes of antibiotics from chemical libraries of >11 million compounds. The working hypothesis is that antibiotic mechanisms of action (MoAs) manifest through physical changes to cellular structures, and that approaches to antibiotic discovery which integrate this MoA information can more reliably discover novel classes of antibiotics than current discovery pipelines. This hypothesis will be tested in two specific aims: (1) develop novel methods of probing and perturbing antibiotic lethality at the single-cell level (Years 1 to 4) and (2) develop a computational platform for deep learning classes of new antibiotics which exploits the mechanisms of action of known antibiotics (Years 2 to 5). During the first phase of this award, microfluidic, optical, and fluorescence microscopy will be used to record changes to physical properties of cytoplasmic and cell envelope components in Escherichia coli cells treated with various bactericidal antibiotics. Additional targeted experiments based on chemical perturbations, physical perturbations, and genetic overexpression and knockout will inform physical and mathematical models, based on multiscale continuum mechanics, that classify the phenotypes associated with antibiotic-induced cell death. During the second phase of this award, a deep learning platform which predicts both antibiotic leads and their predicted MoAs in silico will be developed. This mechanism-guided approach will be used to identify antibiotic leads and their MoAs from vast chemical spaces, and leads will be experimentally validated in vitro against laboratory strains and multidrug-resistant clinical isolates. Leads will be further investigated using human cell cytotoxicity, hit-to-lead, pharmacokinetic, and in vivo mouse bacterial infection experiments. A better understanding of antibiotic MoAs and the development of novel drug discovery efforts with detailed mechanistic underpinnings fit NIH’s public health mission and have direct implications for the prevention and treatment of infectious diseases. This work will establish a quantitative, model-guided platform for better characterizing and discovering antibiotics, one which promises to offer a fertile source of mechanistic information and chemical diversity. By providing the applicant the opportunity to develop his research career, acquire new wet-lab experimental skills, undertake translational coursework, and receive mentorship from leading experts in antibiotics, machine learning, biotechnology, and infectious diseases at MIT, the support and training provided by this award will enable the applicant’s development as an independent researcher.