The emergence of antibiotic resistance poses a global medical threat. The discovery of new antibiotics from naturally-occurring secondary metabolites or from high-throughput screening of chemical libraries has failed to match the pace of resistance mechanisms. Therefore, new approaches for rapidly developing antibiotics are required to address this global medical need. We propose to integrate computational peptide design, deep learning, and high-throughput chemical synthesis to create a general framework for the structure-guided design of new antibiotics. Our recent advances in physics-based and deep-learning-based peptide design algorithms enable the design of structured peptide macrocycles with atomic-level accuracy. We will extend and implement these computational methods to design inhibitors of a validated target for antibiotics — the bacterial type I signal peptidase (SPase I). The secretion and correct folding of critical extracellular and periplasmic proteins in gram-positive and gram-negative bacterial requires the cleavage of their preproteins by SPase I. Naturally-occurring inhibitors of SPase I, arylomycins, show a limited spectrum of activity due to sequence variations in SPase I from different pathogenic bacteria. We propose to implement parallel strategies to design broad-spectrum inhibitors of SPase I. In one strategy, we will leverage the promising interactions between arylomycin and conserved regions of the SPase I active site as starting points and extend these interactions into macrocyclic inhibitors by sampling different backbone conformations and amino acid sequences inside the SPase active site. In a parallel de novo design approach, we will first identify promising docked conformations of individual amino acids to the SPase I pocket and then graft those interactions on pre-enumerated and predesigned sets of millions of structured cyclic peptides. The de novo design approach will be more widely applicable and enable targeting proteins that do not yet have a natural inhibitor identified. We will filter the tens of millions of computational models in silico to find the best design models for experimental validation. The promising design models will be chemically synthesized and screened for binding to SPase I as a massively parallel macrocycle library with tens of thousands of members. We will further test the computationally-designed macrocycles for SPase I binding, inhibition of the proteolytic activity, inhibition of bacterial growth across multiple species, toxicity, and the accuracy of the designed binding mode. The overall workflow will be implemented as iterative cycles of design-synthesis-test-learn to enable simultaneous optimization of potency, bacterial growth inhibition, stability, and other drug-like properties. Overall, SPase I inhibitors developed in this project will be promising candidates for further development as antibiotics. The computational and experimental methods developed during this project will be b...