法医学杂志

法医学杂志 ›› 2025, Vol. 41 ›› Issue (4): 364-370.DOI: 10.12116/j.issn.1004-5619.2025.350703

• 论著快检技术赋能法医毒物学多场景应用专题 • 上一篇    下一篇

基于表面增强拉曼光谱与机器学习的依托咪酯及其结构类似物的快速鉴识

郭紫雯(), 邱天禹, 曹玥()   

  1. 南京医科大学基础医学院法医学系，江苏 南京 211166
  • 收稿日期:2025-06-02 发布日期:2025-11-25 出版日期:2025-08-25
  • 通讯作者: 曹玥
  • 作者简介:郭紫雯（2000—），女，硕士研究生，主要从事法医毒物分析研究；E-mail：guoziwenjp@163.com
  • 基金资助:
    上海市法医学重点实验室暨司法部司法鉴定重点实验室开放课题资助项目(KF202503);国家自然科学基金资助项目(21904068);江苏省科技基金资助项目(BK20201351);教育部“春晖计划”合作科研项目(HZKY20220178)

Rapid Identification of Etomidate and Its Structural Analogues Based on Surface-Enhanced Raman Spectroscopy and Machine Learning

Zi-wen GUO(), Tian-yu QIU, Yue CAO()   

  1. Department of Forensic Medicine, School of Basic Medical Sciences, Nanjing Medical University, Nanjing 211166, China
  • Received:2025-06-02 Online:2025-11-25 Published:2025-08-25
  • Contact: Yue CAO

RichHTML

0

PDF (PC)

5

补充材料

1

摘要：

目的 获取依托咪酯及其结构类似物的差异性光谱特征，建立基于表面增强拉曼光谱（surface-enhanced Raman spectroscopy，SERS）与机器学习算法的快速鉴识方法，用于区分依托咪酯及其结构类似物。 方法 以银纳米颗粒为SERS基底，采集依托咪酯、美托咪酯、丙帕酯和异丙帕酯在1×10-4和1×10-5 mol/L两个浓度点处的SERS光谱，并采集含有1×10-5 mol/L依托咪酯、美托咪酯、丙帕酯和异丙帕酯的血液、尿液样本以及缴获的含依托咪酯的电子烟油的SERS光谱。利用均匀流形逼近与投影（uniform manifold approximation and projection，UMAP）进行非线性降维可视化，并基于XGBoost算法构建分类模型，实现对4种结构高度相似化合物的判别分析。 结果 拉曼光谱在1 398~811 cm-1区间内识别出特征峰微小位移（5~3 cm-1），在血清、尿液、电子烟油样本中实现了无需前处理的定性鉴别，经UMAP降维后不同物质呈现明显聚类分离。XGBoost模型在测试集上实现100%分类准确率，特征权重分析结果表明C-N伸缩振动（841 cm-1）、C=O伸缩振动（1 367 cm-1）和C-O-C反对称振动（1 049 cm-1）为判别关键谱带。 结论 SERS技术结合机器学习可有效放大分子结构细微差异，实现依托咪酯及其结构类似物的快速、准确鉴别，适用于法医毒物鉴定的现场快速筛查。

关键词: 法医学, 毒物分析, 依托咪酯, 结构类似物, 银纳米颗粒, 表面增强拉曼光谱, 机器学习, 快速筛查

Abstract:

Objective To obtain differential spectral characteristics of etomidate and its structural analogues, and to establish a rapid identification method using surface-enhanced Raman spectroscopy （SERS） combined with machine learning algorithms for distinguishing etomidate and its analogues. Methods Silver nanoparticles （AgNPs） were used as the SERS substrate to collect SERS spectra of etomidate, metomidate, propoxate, and isopropoxate at two concentrations of 1×10-4 and 1×10-5 mol/L. SERS spectra were also obtained from blood and urine samples containing 1×10-5 mol/L of etomidate, metomidate, propoxate, and isopropoxate, as well as from confiscated e-cigarette oil containing etomidate. Uniform manifold approximation and projection （UMAP） was employed for nonlinear dimensiona-lity reduction and visualization, and a classification model based on the XGBoost algorithm was constructed to enable discriminant analysis of these four structurally highly similar compounds. Results Minor characteristic peak shifts （5-3 cm-1） were identified in the range of 1 398-811 cm-1. Qualitative identification of the compounds in serum, urine and e-cigarette oil samples was achieved without pretreatment. After UMAP dimensionality reduction, distinct clustering separation among different substances was observed. The XGBoost model achieved 100% classification accuracy on the test set. Feature weight analysis revealed that C-N stretching vibration （841 cm-1）, C=O stretching vibration （1 367 cm-1）, and C-O-C asymmetric vibration （1 049 cm-1） were the key spectral bands for discrimination. Conclusion The combination of SERS and machine learning can effectively amplify subtle differences in molecular structures, enabling rapid and accurate identification of etomidate and its analogues. This approach is suitable for on-site rapid screening in forensic toxicology.

Key words: forensic medicine, toxicological analysis, etomidate, structural analogues, silver nanoparticles （AgNPs）, surface-enhanced Raman spectroscopy （SERS）, machine learning, rapid screening

中图分类号: