Directed evolution of antimicrobial peptides using multi-objective zeroth-order optimization.

Area of Science:

• Biochemistry and computational chemistry
• Drug discovery and development
• Bioinformatics and machine learning

Background:

• Antimicrobial peptides (AMPs) show therapeutic promise but require optimization for enhanced activity and reduced toxicity.
• Existing optimization algorithms for AMPs often suffer from low efficiency, limited diversity, and local optima due to discrete sequence space exploration.

Purpose of the Study:

• To introduce PepZOO, a novel directed evolution method for optimizing multi-properties of AMPs.
• To address the limitations of previous methods by operating in a continuous representation space.

Main Methods:

• PepZOO utilizes a variational autoencoder to project AMPs into a continuous latent space.
• Multi-objective zeroth-order optimization guides the iterative updating of latent embeddings for AMP optimization.
• Molecular docking and dynamics simulations validate the optimized AMPs.

Main Results:

• PepZOO demonstrates superior performance over state-of-the-art methods in enhancing AMPs' antimicrobial function, activity, toxicity profiles, and target binding affinity.
• The method successfully identifies key motifs crucial for maintaining specific peptide properties during evolution.
• PepZOO accelerates the discovery of potential therapeutic peptides by focusing on property optimization.

Conclusions:

• PepZOO offers a new paradigm for AMP optimization by shifting focus from sequence mutation to property exploration.
• This approach enhances efficiency and diversity in discovering novel therapeutic peptides.
• The method validates the effectiveness of continuous representation and multi-objective optimization for AMP design.