An explainable few-shot learning model for the directed evolution of antimicrobial peptides

Area of Science:

• Biochemistry
• Molecular Biology
• Drug Discovery

Background:

• Antibiotic resistance in Gram-negative pathogens is a critical global health threat.
• Traditional antibiotics face limitations, necessitating the development of novel antimicrobial agents.
• Antimicrobial peptides (AMPs) offer a promising alternative due to their unique mechanisms of action.

Purpose of the Study:

• To explore the chemical space of AMPs using deep learning-guided directed evolution.
• To engineer structural modifications of the lipopolysaccharide-binding domain (LBD) for enhanced antimicrobial activity.
• To elucidate the bactericidal mechanisms and directed evolution pathways of engineered AMPs.

Main Methods:

• Utilized a fine-tuned protein language model for few-shot learning on a small dataset.
• Employed deep learning-guided directed evolution to modify the LBD from Marsupenaeus japonicus.
• Performed molecular dynamics simulations and utilized the ladderpath framework to analyze evolution pathways and mechanisms.

Main Results:

• Engineered LBDs exhibited significant antimicrobial activity against various Gram-negative pathogens.
• Identified specific structural modifications conferring potent bactericidal properties.
• Elucidated the mechanism of action and mapped the evolutionary trajectory of AMP development.

Conclusions:

• Deep learning-guided directed evolution is an effective strategy for rational AMP design.
• Explainable few-shot learning facilitates the discovery of novel antimicrobial agents.
• Engineered AMPs hold potential as alternatives to conventional antibiotics for combating resistant Gram-negative infections.