Machine Learning-Assisted Prediction and Generation of Antimicrobial Peptides.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Antibiotic resistance is a growing global health threat, necessitating new therapeutic strategies.
• Antimicrobial peptides (AMPs) show promise due to broad-spectrum activity and low resistance development.
• Accelerating the discovery of novel AMPs is crucial for combating resistant pathogens.

Purpose of the Study:

• To develop and refine a predictive and generative algorithm for discovering novel antimicrobial peptides (AMPs).
• To enhance the efficiency and accuracy of identifying functional AMPs against multidrug-resistant pathogens.

Main Methods:

• Constructed a predictive and generative algorithm incorporating machine learning (ML) for AMP discovery.
• Utilized an eXtreme Gradient Boosting model achieving ≈87% accuracy in AMP/non-AMP classification.
• Experimentally validated generated peptide sequences and refined the algorithm by analyzing physicochemical properties (charge, hydrophobicity).

Main Results:

• The initial algorithm achieved ≈60% experimental accuracy, which was improved to ≈80% after refinement.
• Generated peptides demonstrated activity against bacterial pathogens and fungal strains.
• Minimal off-target toxicity was observed for the validated AMPs.

Conclusions:

• In silico predictive and generative models are powerful tools for discovering and engineering effective AMPs.
• Algorithm refinement based on physicochemical properties significantly enhances experimental validation accuracy.
• This approach offers a viable strategy for developing new treatments against multidrug-resistant pathogens.