Generative Deep Learning Pipeline Yields Potent Gram-Negative Antibiotics

The escalating crisis of multiresistant bacteria demands the rapid discovery of novel Antibiotics that transcend the limitations imposed by the biased chemical space of current libraries. To address this challenge, we introduce an innovative deep learning-driven pipeline for de novo Antibiotic design. Our unique approach leverages a chemical language model to generate structurally unprecedented Antibiotic candidates. The model was trained on a diverse chemical space of drug-like molecules and natural products. We then applied transfer learning using a data set of diverse Antibiotic scaffolds to refine its generative capabilities. Using predictive modeling and expert curation, we prioritized the most promising compounds for synthesis. This pipeline identified a lead candidate with potent activity against methicillin-resistant Staphylococcus aureus. We then performed iterative refinement by synthesizing 40 derivatives of the lead compound. This effort produced a suite of active compounds, with 30 showing activity against S. aureus and 17 against Escherichia coli. Among these, lead compound D8 exhibited remarkable submicromolar and single-digit micromolar potency against the aforementioned pathogens, respectively. Mechanistic investigations point to the reductive generation of reactive species as its primary mode of action. This work validates a deep-learning pipeline that explores chemical space to generate Antibiotic candidates. This process yields a potent nitrofuran derivative and a set of experimentally validated scaffolds to seed future Antibiotic development.

Gram-negative; MRSA; antibiotics; automated synthesis; de novo drug design; deep learning; drug discovery; machine learning.