A98-08 Artificial Intelligence-assisted Design of Novel Antimicrobial Peptides Against Mycobacterium Abscessus

Rationale Antimicrobial drug resistance (AMR) is a critical threat to global health, underscoring the urgent need for novel antibiotic discovery. The persistent rise in AMR infections, exacerbated by antibiotic overuse and the slowdown in new drug discovery, highlights an urgent need for innovative therapeutic strategies. Antimicrobial peptides (AMPs), naturally occurring components of host defense systems, possess broad-spectrum bactericidal activity, making them attractive candidates for combating resistant pathogens. However, their clinical translation remains limited due to issues of cytotoxicity and production cost. Advances in artificial intelligence (AI) now offer a means to overcome these barriers by enabling rational peptide design. Methods We applied an AI-based convolutional neural network trained on AMP structure and minimum inhibitory concentration (MIC) to design synthetic AMPs (SAMPs) predicted to possess antibacterial activity across multiple pathogens, including Mycobacterium species. This model generated 523 novel SAMPs to target ten clinically relevant AMR pathogens (Figure). A subset of 108 SAMPs with predicted antimycobacterial activity was selected for experimental validation. Mycobacterium abscessus (Mab), a rapidly growing, multidrug-resistant non-tuberculous mycobacterium that causes chronic pulmonary infections. Clinical Mab isolates were first characterized for clarithromycin resistance profiles to identify suitable strains for testing. The selected peptides were synthesized and screened in microdilution assays against both clarithromycin-sensitive (MAB1050) and clarithromycin-resistant (MAB391) strains to assess their MICs and bactericidal potential. Peptides demonstrating initial activity were resynthesized at high purity for validation, and hemolytic assays were conducted to evaluate cytotoxicity and potential therapeutic index. Results Our screening identified 52 active AMPs, 36 novel SAMPs and 16 reference peptides, with measurable activity against one or both Mab strains. Among these, two AI-designed SAMPs, peptide A and peptide B, demonstrated consistent bactericidal activity, with peptide B showing MICs of 32 μg/mL against both sensitive and resistant isolates. Importantly, peptide B exhibited minimal hemolytic activity (0.35% at 38 μg/mL), indicating low host cell toxicity. Comparative clustering revealed that active peptides formed distinct sequence groups associated with bacterial specificity, underscoring the AI model’s capacity to generate targeted antimicrobial profiles. We are currently assessing the in vivo efficacy of the lead SAMP candidates using a murine model of pulmonary Mab infection. Conclusions These findings demonstrate that AI-guided peptide design can accelerate the discovery of novel, selective, and safe antimicrobial candidates. Peptide B emerges as a promising therapeutic lead against M. abscessus, and this study highlights how integrating AI and peptide engineering can transform antimicrobial discovery to address the escalating AMR crisis. This abstract is funded by: NIH