Pepxml: ESM2-based extreme multilabel classification of pathogen-targeted antimicrobial peptides.

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Antimicrobial peptides (AMPs) show promise as peptide antibiotics due to broad-spectrum activity and specificity.
• Current AMP prediction methods primarily focus on general functions (antibacterial, antiviral, anticancer), neglecting specific pathogen targeting.
• A significant gap exists in identifying AMPs effective against specific pathogens, complicated by data sparsity and label imbalance.

Purpose of the Study:

• To develop PepXML, a large language model-based tool for extreme multilabel classification of pathogen-targeted AMPs.
• To address the challenge of predicting specific pathogen targets for AMPs.
• To provide a valuable resource for advancing peptide-based therapeutics.

Main Methods:

• Constructed a benchmark dataset of AMPs and their targeted pathogens from public databases.
• Utilized ESM2 for peptide embedding.
• Employed label co-occurrence graph clustering and hard negative sampling to handle data sparsity and imbalance.
• Validated predictions using molecular docking and molecular dynamics simulations.

Main Results:

• Developed PepXML, a novel tool for predicting pathogen-specific antimicrobial peptides.
• Successfully addressed data sparsity and label imbalance challenges in AMP prediction.
• Molecular simulations confirmed the reliability of predictive results and elucidated interaction mechanisms.

Conclusions:

• PepXML offers a robust solution for identifying pathogen-targeted AMPs.
• The tool and its underlying methods advance the field of antimicrobial peptide discovery.
• PepXML is expected to accelerate the development of novel peptide-based therapeutics.