Dual AI Framework Designs Better C-Amidated Antimicrobial Peptides with Interpretable Rules

The researchers developed two complementary models: (1) an Explainable Boosting Machine (EBM) for interpretable design rule extraction from C-amidated AMP sequences, and (2) a fine-tuned ESM2 protein language model for high-accuracy activity prediction. Models were trained on datasets of C-amidated AMPs with known activity against E. coli. Design rules were cross-referenced with published alanine-scanning mutagenesis data. The framework was validated by designing and predicting activity for pardaxin variants.

C-terminal amidation is one of the most common modifications in natural and therapeutic peptides, yet most AMP prediction tools ignore it. By building the first framework specifically for amidated peptides, this study fills a critical gap in computational peptide design. The interpretable component is especially valuable — unlike black-box AI that just gives predictions, this system explains why certain sequences work, enabling researchers to rationally design better peptides.

As AI-driven drug design accelerates, the tension between prediction accuracy (deep learning excels) and interpretability (classical ML excels) is a central challenge. This dual-model approach elegantly resolves it — using the interpretable model for design insights and the deep learning model for deployment predictions. The focus on a specific chemical modification (C-amidation) rather than generic peptide prediction represents a maturing of the field toward clinically relevant peptide design.

The framework was demonstrated against E. coli only — activity predictions for other pathogens would need separate models. The improved pardaxin variant was identified computationally; it's unclear from the abstract whether it was experimentally validated. The training data for C-amidated AMPs may be limited compared to general AMP databases. The framework focuses on one modification type (C-amidation) and doesn't address other common modifications like cyclization or D-amino acid substitutions.