基于SERS和深度学习算法的粪肠球菌检测研究  

Detection and analysis of Enterococcus faecalis based on SERS and deep learning methods

作　　者：刘刚 沈经天 蒋诗妍 黄靖 陈帅[3] 黄祖芳 Gang Liu;Jingtian Shen;hiyan Jiang;Jing Huang;Shuai Chen;Zufang Huang(Fujian Normal University,School of Optoelectronics and Information Engineering,Fuzhou 350117,China;Fujian University of Medicine,School of Stomatology,Fuzhou 350004,China;Fujian University of Medicine,Fujian Medical University School of Stomatology·Affiliated Stomatological Hospital,Fuzhou 350002,China)

机构地区：[1]福建师范大学光电与信息工程学院,福州350117 [2]福建医科大学口腔医学院,福州350004 [3]福建医科大学,福建医科大学口腔医学院·附属口腔医院,福州350002

出　　处：《福光技术》2024年第1期38-44,60,共8页FUJIAN OPTICAL TECHNOLOGY

摘　　要：人体益生菌对于维护消化系统和整体健康至关重要。传统上,细菌的鉴定方法可分为以下三种:检测细菌或其抗原、检测抗体、以及检测细菌的遗传物质。然而,上述方法涉及复杂的实验操作,依赖于特定的实验环境和仪器设备。拉曼光谱技术具备非破坏性、快速、无需样品特殊准备等诸多优势,提供了微生物检测的重要分析手段。在本研究中,首先获得了粪肠球菌的表面增强拉曼光谱(Surface-enhanced Raman Scatting,SERS数据,拉曼光谱在660cm^(-1)、733cm^(-1)和1332cm^(-1)处显示出归属于粪肠球菌的蛋白质,核酸成分的特征性SERS拉曼谱峰。随后,制备了浓度105CFU/mL至1010CFU/mL的六种粪肠球菌样本,开展了粪肠球菌的SERS定量检测分析。此外,利用条件表格生成对抗网络扩增数据并融合原始数据形成混合数据集,并使用一种改进的卷积神经网络模型进行分类。基于改进的卷积神经网络残差网络(Resident Network,ResNet)模型在混合数据集上的准确率达到100%。研究表明,将SERS技术与深度学习方法相结合,有望提供细菌定量分析的新方法。Human probiotics play a crucial role in maintaining the digestive system and overall health.Traditionally,bacterial identification methods can be classified into three categories:detecting bacteria or their antigens,detecting antibodies,and detecting the genetic material of bacteria.However,these methods involve complex experimental procedures,relying on specific laboratory environments and instrumentation.With its non-destructive,rapid,and sample-preparation-free advantages,Raman spectroscopy technology provides a significant analytical tool for microbial detection.In this study,Surface-Enhanced Raman spectroscopic data of Enterococcus faecium were initially obtained,revealing protein and nucleic acid components at 660 cm^(-1),733 cm^(-1),and 1332 cm^(-1)associated with Enterococcus faecium.Subsequently,samples of six Enterococcus faecium strains with concentrations ranging from 105 CFU/mL to 101o CFU/mL were prepared,and quantitative SERS detection analysis of Enterococcus faecium was conducted.Furthermore,a conditional table was generated to augment the data,which was then integrated with the original dataset to form a hybrid dataset.An improved Convolutional Neural Network(CNN)model was employed for classification.The accuracy of the enhanced CNN ResNet model on the hybrid dataset reached 100%.The study indicates that integrating SERS and deep learning methods holds promise for providing a novel approach to bacterial quantitative analysis.

关 键 词：粪肠球菌 拉曼光谱 SERS 卷积神经网络 定量检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]