Scholarlink

<p>Antimicrobial resistance is accelerating worldwide. Antimicrobial peptides (AMPs), owing to their broad-spectrum activity, multimodal mechanisms, and relatively low tendency to induce resistance, are attracting increasing attention. The discovery and design of AMPs has shifted from chance discoveries in natural sources and template-based sequence modification to an AI-driven, data-centric paradigm. This review synthesizes advances across three pillars: (1) AMP datasets and machine-/deep-learning predictors; (2) mining-based discovery from genomes, metagenomes, transcriptomes, proteomes, and exhaustive combinatorial sequence libraries; (3) generation-based design, including the increasingly common predictor-generator frameworks for goal-directed sequence creation. We conclude by outlining current challenges and future opportunities, such as standardized datasets and benchmarks, multi-objective optimization (balancing potency, spectrum, toxicity, and synergy), and tighter integration of AI with automated experimentation. Collectively, AI integration is transforming AMP research from empirical search into a systematic, scalable engineering discipline, accelerating the discovery of potent, low-toxicity, and functionally versatile AMPs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-driven discovery and design of antimicrobial peptides

  • 期刊论文

  • 发表日期：

    2025年11月6日

查看全文
  • Xintian Shen,
  • Youxiang Wang,
  • Peng Zhang,
  • Jian Ji

摘要

Antimicrobial resistance is accelerating worldwide. Antimicrobial peptides (AMPs), owing to their broad-spectrum activity, multimodal mechanisms, and relatively low tendency to induce resistance, are attracting increasing attention. The discovery and design of AMPs has shifted from chance discoveries in natural sources and template-based sequence modification to an AI-driven, data-centric paradigm. This review synthesizes advances across three pillars: (1) AMP datasets and machine-/deep-learning predictors; (2) mining-based discovery from genomes, metagenomes, transcriptomes, proteomes, and exhaustive combinatorial sequence libraries; (3) generation-based design, including the increasingly common predictor-generator frameworks for goal-directed sequence creation. We conclude by outlining current challenges and future opportunities, such as standardized datasets and benchmarks, multi-objective optimization (balancing potency, spectrum, toxicity, and synergy), and tighter integration of AI with automated experimentation. Collectively, AI integration is transforming AMP research from empirical search into a systematic, scalable engineering discipline, accelerating the discovery of potent, low-toxicity, and functionally versatile AMPs.