Deep Learning-Based Prediction of Protein Structure for Novel Antimicrobial Peptide Design

Abstract

Background: Rising antimicrobial resistance has renewed interest in antimicrobial peptides (AMPs) as anti-infective candidates,but rational design remains constrained by limited structural insight. Objective: This narrative review synthesizes publishedliterature on how deep learning-based structure prediction and generative sequence design are being applied to novel AMPdiscovery. Methods: We examine peer-reviewed studies on structure predictors (AlphaFold, RoseTTAFold, ESMFold), generativediffusion and sequence-design tools (RFdiffusion, ProteinMPNN), deep generative AMP models (HydrAMP, AMPGAN v2), andlarge-scale machine-learning mining pipelines validated against curated databases (APD3, DBAASP, DRAMP). Findings: Structurepredictors offer rapid but variably accurate models for short, often disordered peptides, while generative and mining pipelines haveproduced experimentally validated hits, including hundreds of thousands of computationally proposed candidates yielding dozensto hundreds of confirmed active peptides across independent studies. Conclusions: Deep learning has substantially accelerated AMPcandidate generation and prioritization, but heterogeneous training data, peptide-scale structural uncertainty, and unresolvedtoxicity prediction remain key barriers to clinical translation.