New article in Nature Communications: Generative AI Discovers Potent, Low-Toxicity Antimicrobial Peptides

Konstantinos Markakis, Shanghyeon Kim, Cheng-En Tan, and Prof. Ilias Tagkopoulos published Discovery of potent low-toxicity antimicrobial peptides through diffusion modeling in Nature Communications on 7/7/2026.

About this work: Can generative AI help discover new antimicrobial peptides that combine strong antibacterial activity with low toxicity? This work introduces ARCADIAMP, an AI-assisted platform that combines diffusion-based peptide generation with antibacterial activity prediction and sequence-novelty screening. From 1,000,000 generated sequences, the pipeline identified 76 candidates that passed its strong-activity and novelty filters, selected ten for synthesis, and found that eight achieved MICs ≤ 32 μg/mL in experimental testing. The lead peptide, Arcinin, showed MICs of 8–32 μg/mL against the tested pathogens, low hemolytic activity with an LC₅₀ > 512 μg/mL in human red blood cells, and retained an MIC of 32 μg/mL in 50% bovine serum for four tested species. In a mouse wound-infection model, Arcinin reduced viable S. aureus and E. coli bacterial counts by 4.56 and 4.47 log₁₀ CFU, respectively; treated wounds also closed significantly faster than PBS controls and showed enhanced re-epithelialization. Together, the results demonstrate how AI-guided peptide generation and screening can move from computational design to experimentally validated antimicrobial candidates.

Reference: Markakis, Konstantinos, Shanghyeon Kim, Cheng-En Tan, and Ilias Tagkopoulos. “Discovery of potent low-toxicity antimicrobial peptides through diffusion modeling.” Nature Communications 17, 8396 (2026). doi: 10.1038/s41467-026-75030-8