Designing Novel Antimicrobial Peptides Against Acinetobacter baumannii Using Artificial Intelligence

This study aims to design novel antimicrobial peptides (AMPs) targeting Acinetobacter baumannii, a multidrug-resistant ESKAPE pathogen associated with high mortality rates among vulnerable hospital populations, particularly those in intensive care units. We tested the hypothesis that substituting polar uncharged and special-case non-polar neutral amino acids with positively charged or hydrophobic residues can improve AMP probability, reduce predicted toxicity, and enhance efficacy. Using peptide databases and artificial intelligence-based computational tools—including GRAMPA, CAMPR3, AlphaFold, APD3, and DBAASP—we applied an exhaustive substitution strategy, generating 81 sequence variants (3⁴ combinations at four targeted positions) and predicting their key characteristics. The peptide variants with the highest predicted AMP probability and lowest toxicity were then further optimized using ToxinPred3.0, a machine learning-based toxicity prediction and simulation tool, via a single-residue substitution to reduce predicted toxicity. The final optimized peptide exhibited a high AMP score, zero predicted toxicity, and favorable physicochemical properties, suggesting enhanced targeting and selectivity against A. baumannii. This study highlights the utility of computational approaches in assisting the rational design of AMPs, providing a cost-effective and time-efficient strategy for developing alternatives to conventional antibiotics.