Discovery of naturally inspired antimicrobial peptides using deep learning

Area of Science:

• Microbiology and Bioinformatics
• Drug Discovery
• Synthetic Chemistry

Background:

• Non-ribosomal peptides (NRPs) are crucial for novel antibiotic development.
• Mining silent microbial NRPS gene clusters aids in discovering bioactive peptides.
• Bioinformatic and deep learning approaches accelerate natural product discovery.

Purpose of the Study:

• To efficiently discover and evaluate novel antimicrobial peptides from bacterial NRPS gene clusters.
• To utilize deep learning for identifying promising peptide scaffolds and optimizing lead compounds.
• To explore the therapeutic potential of newly discovered antimicrobial peptides.

Main Methods:

• Genome mining of 216,408 bacterial genomes to identify NRPS gene clusters.
• Dereplication of identified clusters to obtain unique peptide scaffolds.
• Deep learning-based scoring for candidate selection, followed by solid-phase synthesis and antibacterial evaluation.
• Amino acid optimization of lead peptides guided by deep learning algorithms.

Main Results:

• Identified 335,024 NRPS gene clusters, yielding 328 unique peptide scaffolds.
• Synthesized five antimicrobial peptide candidates (P1-P5); P2 showed potent activity (MIC50: 1-2 μM).
• Optimized P2 to P2.2, demonstrating significantly enhanced antibacterial activity, membrane disruption, and synergistic effects with conventional antibiotics.

Conclusions:

• Deep learning effectively accelerates the discovery of naturally inspired antimicrobial peptides from silent biosynthetic gene clusters.
• The optimized peptide P2.2 exhibits promising therapeutic potential due to enhanced activity, reduced toxicity, and synergistic properties.
• This study highlights a powerful strategy for identifying novel antibiotic leads from microbial genomic data.