Discovery of potent low-toxicity antimicrobial peptides through diffusion modeling

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• The rise of multidrug-resistant bacteria necessitates novel antimicrobial discovery platforms.
• Existing screening methods face challenges in efficiency and identifying candidates with multifaceted properties.

Purpose of the Study:

• To introduce ARCADIAMP, an AI-driven platform for generative and virtual screening of antimicrobial peptides (AMPs).
• To optimize AMP discovery by integrating iterative learning, discrete denoising diffusion models, and ESM2-based classification for activity, toxicity, and stability.

Main Methods:

• Coupling a discrete denoising diffusion probabilistic model with a two-stage Evolutionary Scale Modeling 2 (ESM2)-based classifier.
• Iterative generation, classification, and prioritization of potential antimicrobial peptides.
• Experimental validation including MIC assays, hemolytic activity tests, serum stability assays, and mechanistic studies (microscopy, depolarization, time-kill kinetics, molecular dynamics).

Main Results:

• Eight of ten screened peptide candidates exhibited antimicrobial activity (MIC ≤ 32 μg/mL).
• Arcinin, a generated candidate, displayed potent activity against ESKAPE pathogens (MIC 8-32 μg/mL), low hemolytic activity (LC50 > 512 μg/mL), and retained activity in serum.
• Arcinin demonstrated AMP-like membrane interaction mechanisms and achieved a 4-log reduction in bacterial burden in a murine wound model, promoting wound healing.

Conclusions:

• ARCADIAMP effectively generates, classifies, and prioritizes AMPs with desirable therapeutic properties.
• Arcinin represents a promising therapeutic candidate for treating bacterial infections, demonstrating efficacy in vitro, in vivo, and mechanistic validation.
• The AI-assisted iterative optimization framework offers a scalable approach for discovering novel biologics.