A Dual-Model Machine Learning Framework for Interpretable Design and Ensemble Prediction of C-Amidated Antimicrobial...

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) are vital alternatives to conventional antibiotics.
• Chemical modifications, such as C-terminal amidation, significantly enhance AMP efficacy, stability, and resistance.
• Existing predictive models often overlook these crucial modifications.

Purpose of the Study:

• To develop an integrated framework for designing and predicting C-terminal amidated AMPs.
• To target specific bacterial pathogens, exemplified by Escherichia coli.
• To improve the real-world applicability of AMPs through advanced computational methods.

Main Methods:

• Integrated framework combining an interpretable Explainable Boosting Machine (EBM) for design rules and a fine-tuned ESM2 deep learning model for prediction.
• Development of a computational tool, CAmidPred, for predictive classification and amino acid pattern analysis.
• Validation against published alanine-scanning experiments to ensure reliability.

Main Results:

• CAmidPred successfully extracts actionable sequence-level design rules for C-terminal amidated AMPs.
• The framework identified a pardaxin variant with enhanced activity against Escherichia coli.
• Demonstrated the practical utility of the dual-model approach in targeted AMP design.

Conclusions:

• The integrated dual-model framework offers a powerful approach for designing effective C-terminal amidated AMPs.
• CAmidPred provides valuable insights into sequence-activity relationships, facilitating rational drug design.
• This methodology advances the development of novel antimicrobial agents to combat bacterial infections.