AI-driven discovery of self-assembled antimicrobial peptides with dual antibacterial and wound-healing functions against multidrug-resistant infections

The rapid emergence of multidrug-resistant bacterial infections, particularly those caused by methicillin-resistant Staphylococcus aureus and carbapenem-resistant Acinetobacter baumannii , poses a critical challenge to global healthcare. Antimicrobial peptides (AMPs) are promising alternatives, yet their design is challenged by vast sequence space and the need to balance potency with safety. This study establishes an deep learning model AMP-SAGE to achieve intelligent screening of a virtual decapeptide library composed of four amino acids (lysine (K), tryptophan (W), phenylalanine (F), and arginine (R)), and further mines for peptides with self-assembly potential that feature an FF self-assembly motif at the core flanked by R-W-K residues, thereby enabling the de novo discovery of self-assembling AMPs. Ten candidate peptides with predicted antimicrobial probability scores≥0.5 and self-assembly potential are randomly screened. After systematic biological evaluation, R2 and K4 are selected for further investigation. The candidate peptides R2 and K4 show broad-spectrum activity (minimum inhibitory concentration (MIC): 1–32 μM) and achieve>99.99% bacterial killing within 1–2 h. Spectroscopic characterization reveals distinct self-assembly behaviors. They operate via multimodal mechanisms, including membrane disruption and reactive oxygen species (ROS) induction, and exhibit high biosafety (hemolysis TI: 150.55 for R2, 1505.75 for K4) and stability. In murine wound models, both peptides significantly reduce bacterial loads, accelerate healing, and promote organized collagen deposition. This work establishes an AI-accelerated pipeline for developing self-assembled AMPs with combined antimicrobial and wound-healing functions against multidrug-resistant infections.