Speaker: Jason Yang, Rutgers University

Title: Interpretable machine learning insights into the coupling between antibiotic lethality and bacterial metabolism

Abstract: Antibiotic stress induces significant metabolic remodeling in bacteria to fundamental processes such as central carbon metabolism as part of their lethality. However, the mechanisms driving such changes are poorly understood as it is difficult to decipher which metabolic pathways are relevant to antibiotic killing and how they are coupled. To address these, we developed a “white-box machine learning” approach integrating biochemical screening with metabolic network modeling and machine learning to rapidly identify metabolic pathways capable of altering antibiotic lethality. Applying this approach to three different bactericidal antibiotics in E. coli, we discovered the novel contribution of purine biosynthesis to antibiotic-induced death physiology. Model analysis revealed that purine biosynthesis imposes a significant burden on ATP utilization and demand, creating a large driving force for central carbon metabolism activity. Using a combination of genetic, biochemical, and phenotypic assays, we demonstrated that purine biosynthesis activity significantly contributes to both antibiotic lethality and antibiotic-induced changes in central metabolism through its effects on ATP demand. We further demonstrated that pyrimidine biosynthesis exerts opposite effects via network crosstalk with purine biosynthesis.