Machine Learning Accelerated Discovery of Antimicrobial Inorganic Nanomaterials

The growing prevalence of infectious diseases and the increasing threat of bacterial resistance have drawn widespread attention to antimicrobial inorganic nanomaterials. However, the diversity, abundance, and complex mechanisms of these materials present significant challenges in identifying new agents that are both efficient and cost-effective with broad-spectrum activity. In response, this study applied machine learning for the first time to discover antimicrobial inorganic nanomaterials. Information on over 2,000 antimicrobial nanomaterials was extracted from more than 8,000 papers. An unsupervised machine learning analysis was conducted to assess data distribution and explore the relationships between material features and antimicrobial activity in high-dimensional space. A series of machine learning models were trained. Through the evaluation of six performance metrics, five key features were identified from 27 dimensions. To further quantify the structure–activity relationships, a genetic programming-symbolic classification model was employed to generate a precise mathematical formula with a prediction accuracy of 0.83. Using this formula, 43 new antimicrobial inorganic nanomaterials were predicted. Of these, four nanomaterials were synthesized and their antibacterial properties were experimentally validated. This work not only provides a next generation approach for designing antimicrobial inorganic nanomaterials but also opens new avenues for applying machine learning in materials science.传染性疾病的日益流行与细菌耐药性威胁的不断加剧，使得抗菌无机纳米材料（antimicrobial inorganic nanomaterials）受到了学界广泛关注。然而这类材料的多样性、丰度与复杂作用机制，为筛选兼具高效性、成本效益与广谱活性的新型抗菌剂带来了显著挑战。为此，本研究首次将机器学习（machine learning）应用于抗菌无机纳米材料的发现工作。研究从8000余篇学术文献中提取了2000余种抗菌纳米材料的相关信息。开展了无监督机器学习分析，以评估数据分布特征，并探究高维空间中材料属性与抗菌活性之间的内在关联。训练了一系列机器学习模型，通过六项性能指标的综合评估，从27个特征维度中筛选出5个关键特征。为进一步量化构效关系（structure–activity relationships），本研究采用了遗传编程-符号分类模型（genetic programming-symbolic classification model），生成了预测精度达0.83的精准数学公式。借助该公式，研究预测得到43种新型抗菌无机纳米材料，其中4种已完成合成，并通过实验验证了其抗菌性能。本研究不仅为抗菌无机纳米材料的设计提供了新一代技术路径，也为机器学习在材料科学领域的应用开辟了全新方向。