Controllable Generation of Pathogen-Specific Antimicrobial Peptides Through Knowledge-Aware Prompt Diffusion Model.

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Generative models show potential in antimicrobial peptide (AMP) design.
• Limited data for specific pathogens hinders the development of targeted AMPs.
• Existing methods struggle with generating pathogen-specific peptides.

Purpose of the Study:

• Introduce KPPepGen, a controllable generative framework for designing pathogen-specific AMPs.
• Extend KPPepGen for peptide optimization using prompt-guided diffusion and combinatorial mutations.
• Address data scarcity challenges in AMP design for targeted pathogens.

Main Methods:

• Developed KPPepGen, a framework using knowledge-aware pathogen prompts from Gene Ontology and knowledge graphs.
• Integrated prompt-guided partial diffusion with multi-site combinatorial mutations for peptide optimization.
• Utilized a diffusion model guided by knowledge injections for biologically plausible peptide generation.

Main Results:

• KPPepGen generated valid peptides for 56 pathogens, showing high novelty and favorable properties.
• Achieved over 10% performance improvement for pathogens with limited training data.
• Generated nine novel peptides with strong antimicrobial activity and low cytotoxicity against E. coli and S. aureus in wet-lab evaluations.
• Demonstrated high success rate (44.3%) in Magainin 2 optimization, with a 7.6% average improvement over ESM-based methods.

Conclusions:

• KPPepGen effectively generates pathogen-specific AMPs, overcoming data scarcity limitations.
• The framework captures essential sequence and structure patterns for individual pathogens.
• KPPepGen significantly enhances AMP performance and offers a promising tool for antimicrobial drug discovery.