Deep-Learning-Assisted Discovery of Marine Antimicrobial Peptides: A Roadmap Integrating Omics Mining, Structure-Guided Diffusion Design, and Sustainable Aquaculture Translation

Antimicrobial peptides (AMPs) offer diverse antibacterial mechanisms, but candidate generation alone does not resolve the linked demands of potency, host safety, stability, manufacturability, and application-specific exposure. This review presents a roadmap that integrates two complementary computational routes for marine AMP discovery. De novo generation expands sequence space, whereas omics mining preserves biological provenance and enables recovery of encoded or encrypted peptides from marine organisms and microbiomes. Structure-guided diffusion is positioned as a selective third route when a defined target, motif, or stable structural template is available, rather than as a universal replacement for sequence-based AMP models. We further examine aquaculture as a translational setting in which efficacy must be evaluated together with salt and feed tolerance, environmental fate, tissue residues, resistance risk, and export-related requirements. Current evidence remains limited at the direct intersection of marine provenance and deep learning, particularly for target-conditioned structural design. We therefore propose an evidence-based workflow that links marine sequence resources, model selection, matched experimental validation, and application-specific deployment. This framework is intended to move marine AMP research from candidate prediction toward reproducible and testable translation.