News

Artificial intelligence is dramatically accelerating the discovery and development of peptide-based drugs, offering new hope in the fight against antimicrobial resistance (AMR). Researchers are leveraging advanced AI models to design novel peptides with potent activity against drug-resistant bacteria, including challenging pathogens like Methicillin-resistant Staphylococcus aureus (MRSA) and Acinetobacter baumannii.

Key Takeaways

• AI-Driven Design: Sophisticated AI algorithms, including machine learning and protein language models, are enabling the rapid design and optimisation of antimicrobial peptides (AMPs).
• Targeting Resistance: These AI-developed peptides show significant promise in combating drug-tolerant bacterial phenotypes such as persisters and biofilms, which are notoriously difficult to treat with conventional antibiotics.
• Enhanced Efficacy and Safety: New peptides are demonstrating potent activity against multi-drug resistant strains, with some showing improved stability and reduced toxicity compared to existing treatments.
• Broad Applicability: The AI methodologies are proving versatile, applicable to designing peptides against various pathogens, including bacteria and fungi.

CAMPER: A Mechanistic AI Framework

A novel AI framework called CAMPER (Constraint-driven AMP Engineering with Ranking) integrates machine learning with biophysically informed scoring functions. This hybrid approach prioritises the design of peptides that specifically target bacterial cell membranes. By combining statistical learning with mechanistic constraints on amphipathicity, helicity, charge, and hydrophobicity, CAMPER has successfully identified potent peptide candidates like WP-CAMPER1. This peptide has shown strong activity against MRSA persisters and biofilms, both in vitro and in vivo, offering a promising new avenue for treating persistent infections.

Few-Shot Learning for Data-Scarce Pathogens

Addressing the challenge of limited data for certain pathogens, such as Acinetobacter baumannii, researchers have developed a few-shot learning-based sequential ensemble model pipeline (FSLSMEP). This approach uses pre-trained protein language models and multiple fine-tuning stages to effectively identify promising antimicrobial peptides even with scarce training data. The FSLSMEP has successfully identified potent peptides, including EME7(7), which demonstrates comparable efficacy to polymyxin B against A. baumannii infections but with significantly reduced kidney toxicity.

De Novo Design for Multifunctional Peptides

Another AI-driven strategy, DLFea4AMPGen, focuses on the de novo design of peptides with multiple bioactivities, including antibacterial, antifungal, and antioxidant properties. By analysing features learned from deep learning models and incorporating insights from amino acid contributions (via SHAP analysis), this method generates peptides with enhanced efficacy and potentially reduced host inflammatory responses. The resulting peptides, such as D1 and D2, have shown broad-spectrum activity against various bacterial strains, including drug-resistant ones, and have demonstrated therapeutic benefits in vivo by reducing bacterial load and alleviating inflammation.

These advancements highlight the transformative potential of AI in accelerating the discovery of next-generation peptide therapeutics, offering critical new tools to combat the escalating threat of antimicrobial resistance.

Sources

• CAMPER: mechanistic artificial intelligence for designing peptides that target MRSA persisters, Nature.
• Uncovering evolutionarily remote and highly potent antimicrobial peptides with protein language models |
  Nature Biomedical Engineering, Nature.
• Discovery of antimicrobial peptides targeting Acinetobacter baumannii via a pre-trained and fine-tuned
  few-shot learning-based pipeline, Nature.
• DLFea4AMPGen de novo design of antimicrobial peptides by integrating features learned from deep learning
  models, Nature.