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

Using cerastecin Cpd 4 as a template, Link-INVENT and AutoMolDesigner were applied for molecular design and chemical derivatization, yielding the MsbA-targeted small molecule Y-11 with an MIC of 0.5 g/mL against A. baumannii. Relative to existing MsbA-targeted molecules that lack sufficient potency or have unfavorable properties, this work expands chemical space through AI-driven design and reports a defined antibacterial activity value. Antibacterial activity is reported as an MIC value, representing in vitro activity evaluation.

Y-11 showed equivalent potency to Cpd4 against carbapenem-resistant A. baumannii, together with lower cytotoxicity, hemolysis, and spontaneous resistance frequency. It improves safety- and resistance-related properties while maintaining potency against resistant strains, offering a more favorable starting point for further development. Based on comparative evaluation of cytotoxicity, hemolysis, and spontaneous resistance frequency, representing in vitro safety and resistance indicators.

In vivo efficacy study demonstrated that Y-11 could effectively reduce bacterial loads in mice infected with A. baumannii. It advances evaluation from in vitro activity to an animal infection model, supporting in vivo effectiveness of the candidate molecule. Based on in vivo efficacy evaluation in a mouse infection model, representing animal-experiment evidence.

Mechanism studies combining molecular dynamics simulation, biochemical assay, and TEM analysis suggested that Y-11 inhibits lipooligosaccharide transport and impairs outer membrane formation, probably by competitively binding the substrate binding site of MsbA and modulating ATPase activity. Beyond activity discovery, it provides an initial mechanistic explanation linking the phenotype to MsbA target function. Supported by multiple approaches including molecular dynamics simulation, biochemical assay, and TEM analysis, representing mechanistic evidence with conclusions stated as probable.

This work addresses the therapeutic need for drug-resistant Gram-negative bacterial infections, using MsbA as the target and cerastecin Cpd 4 as the template, designing and derivatizing Y-11 through AI tools; it applies to in vitro activity evaluation against A. baumannii (including carbapenem-resistant strains) and in vivo efficacy validation in a mouse infection model, with mechanistic interpretation centered on lipooligosaccharide transport and outer membrane formation and conclusions stated as probable. For researchers interested in AI-driven molecular design workflows, it offers a complete example from template selection and tool use to activity and mechanism evaluation; for those focused on antibiotic development, Y-11 can serve as a starting point for further structural optimization and developability studies.

The currently available text is summary-level content lacking figures and supplementary data, so details such as Y-11's chemical structure, specific MIC assay conditions, quantitative cytotoxicity and hemolysis results, spontaneous resistance frequency values, dosing regimens and magnitude of bacterial load changes in the mouse infection model, and specific parameters of molecular dynamics simulation and biochemical assays cannot be confirmed here; the mechanistic conclusion is stated as probable, and the causal relationship between competitive binding to the substrate binding site and modulation of ATPase activity still needs to be checked against the original data.