cTo design and discover new antimicrobial peptides (AMPs) with high levels of antimicrobial activity, a number of machinelearning methods and prediction methods have been developed. Here, we present a new prediction method that can identify novel AMPs that are highly similar in sequence to known peptides but offer improved antimicrobial activity along with lower host cytotoxicity. Using previously generated AMP amino acid substitution data, we developed an amino acid activity contribution matrix that contained an activity contribution value for each amino acid in each position of the model peptide. A series of AMPs were designed with this method. After evaluating the antimicrobial activities of these novel AMPs against both Gram-positive and Gram-negative bacterial strains, DP7 was chosen for further analysis. Compared to the parent peptide HH2, this novel AMP showed broad-spectrum, improved antimicrobial activity, and in a cytotoxicity assay it showed lower toxicity against human cells. The in vivo antimicrobial activity of DP7 was tested in a Staphylococcus aureus infection murine model. When inoculated and treated via intraperitoneal injection, DP7 reduced the bacterial load in the peritoneal lavage solution. Electron microscope imaging and the results indicated disruption of the S. aureus outer membrane by DP7. Our new prediction method can therefore be employed to identify AMPs possessing minor amino acid differences with improved antimicrobial activities, potentially increasing the therapeutic agents available to combat multidrug-resistant infections.A ntimicrobial peptides (AMPs) are produced by multicellular organisms to defend against microbial infections. Along with potent antimicrobial activity, many AMPs also have the ability to enhance immunity by functioning as immunomodulators (1-3). AMPs therefore have excellent therapeutic potential, especially in light of increased drug resistance to many conventional antibiotic therapies. A number of naturally occurring peptides and synthetic derivatives have been developed or are currently in development (2, 4-6). Although AMPs vary in length, amino acid composition, and structure, they share some similarities, such as electrical charge and amphipathicity (3, 7). To determine the characteristics that are important in antimicrobial activity, bioinformatic tools and prediction methods have been developed (8), all based to some extent on the sequence similarities between peptides (9-11).To optimize the antimicrobial activity of identified AMPs and to predict novel peptide sequences, we present a machine-learning method based on the concept of an antimicrobial activity contribution score for each amino acid. Here, we consider that each amino acid in a peptide sequence possesses a different level of importance for the biological activity of that peptide, and this is represented by an assigned score. By calculation of each amino acid's contribution score, we can predict the antimicrobial activity of an AMP.To verify our results, we tested some of our designed AMPs ...