AI-guided discovery and optimization of antimicrobial peptides through species-aware language model

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Antibiotic resistance necessitates novel antimicrobial agents.
• Antimicrobial peptides (AMPs) offer a promising alternative due to diverse mechanisms and lower resistance development.
• Current methods for AMP discovery are often slow and resource-intensive.

Purpose of the Study:

• To develop and validate a target species-aware AI model, LLAMP (Large Language model for AMP activity prediction), for predicting AMP activity.
• To accelerate the discovery of novel AMPs with high potency and selectivity.
• To identify critical amino acid residues influencing AMP efficacy.

Main Methods:

• Leveraged pre-trained language models to create LLAMP for predicting minimum inhibitory concentration (MIC) values.
• Screened approximately 5.5 million peptide sequences using LLAMP.
• Analyzed attention values to identify key amino acid residues and engineered a promising peptide sequence (peptide 13-5).

Main Results:

• LLAMP successfully predicted AMP activity and identified peptides 13 (most selective) and 16 (most potent).
• Critical residues like Trp, Lys, and Phe were identified as important for AMP activity.
• Engineered peptide 13-5 showed enhanced antimicrobial activity and potency comparable to pexiganan, though with reduced selectivity.

Conclusions:

• AI, specifically LLAMP, can significantly expedite the discovery of novel peptide-based antibiotics.
• The identified peptides demonstrate potential for combating antibiotic-resistant bacteria.
• Targeted engineering based on AI-driven insights can optimize AMP properties for therapeutic applications.