tsAMP: a strain-level antimicrobial peptide identification framework based on large language models and pathogen...

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• The rise of multidrug-resistant bacteria necessitates novel therapeutic strategies.
• Antimicrobial peptides (AMPs) show promise as alternatives to conventional antibiotics.
• Accurate computational identification and prediction of AMP activity are crucial.

Purpose of the Study:

• To develop tsAMP, a computational framework for identifying antimicrobial peptides (AMPs).
• To improve the accuracy of strain-level Minimum Inhibitory Concentration (MIC) prediction for AMPs.
• To provide a tool for prioritizing AMP candidates for experimental validation.

Main Methods:

• Integrated the ESM-1v protein language model with multidimensional feature extraction.
• Trained the model on a dataset of AMPs and non-AMP sequences from metagenomes.
• Developed tsAMP for both AMP identification and bacterial inhibition/MIC prediction.

Main Results:

• Achieved an F1-score of 0.958 for AMP identification, surpassing existing tools.
• Maintained F1-scores above 0.8 for bacterial inhibition prediction across 33 species.
• Demonstrated high performance in strain-specific MIC prediction (MSE = 0.214, R^2 = 0.634) and validated predictive reliability against experimental data.

Conclusions:

• tsAMP offers a robust computational framework for identifying and prioritizing AMP candidates.
• The tool facilitates the discovery of novel AMPs to combat antimicrobial resistance.
• tsAMP aids in the downstream experimental characterization of potential antimicrobial agents.