Deep reinforcement learning-driven discovery of a MsbA-targeted small-molecule antibiotic for the treatment of Acinetobacter baumannii infection

Deep reinforcement learning has been instrumental in the discovery of a new small-molecule antibiotic, Y-11, capable of targeting and combating Acinetobacter baumannii infections. MsbA, a crucial protein involved in lipopolysaccharide biosynthesis, was identified as a promising target for developing novel antibiotics. However, existing MsbA-targeted molecules often lacked efficacy or had undesirable properties, prompting the need for an expanded chemical space.

The research team utilized two AI-based tools, Link-INVENT and AutoMolDesigner, to design and optimize molecules based on the template cerastecin Cpd4. Through molecular design, chemical derivatization, and antibacterial activity evaluation, they identified Y-11 as a potent candidate. Y-11 displayed similar potency to Cpd4 against carbapenem-resistant A. baumannii, while exhibiting reduced cytotoxicity, hemolysis, and spontaneous resistance frequency.

In vivo efficacy studies confirmed that Y-11 effectively reduced bacterial loads in mice infected with A. baumannii. Mechanistic studies, comprising molecular dynamics simulations, biochemical assays, and transmission electron microscopy analysis, suggested that Y-11 functions by competitively binding to the substrate binding site of MsbA and modulating its ATPase activity, thereby inhibiting lipooligosaccharide transport and impairing outer membrane formation.

The successful discovery of Y-11 via AI-driven drug design presents a promising foundation for future antibiotic development.