A Multi-Property Optimizing Generative Adversarial Network for de novo Antimicrobial Peptide Design.

Area of Science:

• Biotechnology and Pharmaceutical Sciences
• Computational Chemistry and Drug Design

Background:

• Antimicrobial peptides (AMPs) are vital for novel anti-infective drug development due to broad-spectrum activity and low resistance potential.
• Traditional laboratory synthesis of AMPs is laborious and time-consuming.
• Existing computational methods struggle to optimize multiple AMP properties simultaneously.

Purpose of the Study:

• To introduce a novel computational approach, MPOGAN, for designing antimicrobial peptides (AMPs) with optimized multiple properties.
• To address the limitations of current methods in simultaneously enhancing antimicrobial potency, reducing cytotoxicity, and increasing diversity.

Main Methods:

• Development of a Multi-Property Optimizing Generative Adversarial Network (MPOGAN) model.
• Iterative learning of peptide-property relationships using a dynamically updated dataset.
• Extensive computational testing to evaluate MPOGAN's design capabilities.

Main Results:

• MPOGAN successfully generated AMPs with potent antimicrobial activity, reduced cytotoxicity, and increased diversity.
• Ten designed AMPs were synthesized, with nine demonstrating antimicrobial activity and low cytotoxicity.
• Two synthesized peptides exhibited potent broad-spectrum antimicrobial activity combined with significantly reduced cytotoxicity.

Conclusions:

• MPOGAN offers a superior computational strategy for designing multifunctional antimicrobial peptides.
• The developed AMPs show significant potential for downstream applications in anti-infective therapies.
• This AI-driven approach accelerates the discovery and optimization of next-generation antimicrobial agents.