Antimicrobial peptides (AMPs) have emerged as promising alternatives to conventional antibiotics, addressing the growing concern of antibiotic resistance. Their efficacy primarily relies on two key properties: amphiphilicity and cationic charge, which promote targeted action on bacterial membranes. The Lynronne family, identified in the bovine rumen through metagenomic screening, already exhibits notable antimicrobial activity and low toxicity, making it a strong candidate for further development [1]. Our goal is to enhance the therapeutic potential of these AMPs using conventional and innovative strategies. Traditional approaches include substitutions with cationic amino acids and D/L enantiomeric modifications.

In parallel, we employed artificial intelligence techniques using the MRL (Molecular Reinforcement Learning) model for AMP design, based on the biophysical properties of the Lynronne family (cationic charge, hydrophobicity). The activity and toxicity of the designed peptides were first evaluated in silico using various prediction tools (MIC predictors and toxicity/hemolysis predictor). Their antimicrobial activity against diverse pathogens and their cytotoxicity on human cells were then measured, allowing comparison between predicted and measured activities. Their mechanism of action was studied via biophysical techniques, membrane permeabilization assays, and molecular dynamics simulations (GROMACS), confirming specific interactions with bacterial membranes and membranolytic effects. This study highlights how combining rational design strategies and AI can optimize Lynronne peptides, reinforcing their potential as alternatives to conventional antibiotics.