A generative artificial intelligence approach for peptide antibiotic optimization

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Antibiotic resistance is a growing global health crisis, necessitating novel antimicrobial agents.
• Current artificial intelligence (AI) methods for drug discovery often focus on screening large libraries or broad generation, not optimizing existing peptide structures under specific constraints.

Purpose of the Study:

• To introduce ApexGO, a novel generative AI approach for optimizing peptide scaffolds to accelerate antibiotic discovery.
• To demonstrate ApexGO's capability in designing potent antimicrobial peptides by modifying existing templates.

Main Methods:

• Utilized a transformer variational autoencoder to embed peptide sequences in a latent space.
• Employed Bayesian optimization to efficiently propose sequence modifications for enhanced antimicrobial potency.
• Generated and synthesized optimized peptide derivatives from ten template peptides.

Main Results:

• ApexGO achieved an 85% experimental hit rate and a 72% success rate in enhancing antimicrobial activity against Gram-negative pathogens.
• Synthesized 100 compounds, with optimized derivatives showing improved antimicrobial properties, mechanism of action, and cytotoxicity profiles.
• AI-optimized molecules demonstrated potent anti-infective activity in preclinical models of *Acinetobacter baumannii* infection, outperforming controls.

Conclusions:

• ApexGO represents a significant advancement in generative AI for peptide design and antibiotic optimization.
• The approach offers a powerful tool to accelerate the discovery and development of new antimicrobial therapies.
• ApexGO shows potential for overcoming the challenges posed by rising antibiotic resistance.