PepCL enables accurate antimicrobial peptide identification via contrastive learning

Antimicrobial peptides (AMPs) are promising anti-infective candidates because of their broad antimicrobial potential and favorable resistance-related characteristics. However, existing computational methods for AMP identification pay insufficient attention to the discriminative refinement of residue-level representations. In this study, we present PepCL, a two-stage framework for predicting activities of AMPs. The first stage pretrains residue-level representations by a contrastive learning module to enhance discriminative abilities, and the second stage decodes peptide activity based on the pretrained contrastive features. Evaluated on three types of antimicrobial peptides, PepCL achieves much better performance than state-of-the-art baselines across all three tasks and demonstrates strong interpretability of its learned residue representations across the two stages.