Abstract A method for analysis of various biotoxins and biological modifiers based on electrospray ionization-ion mobility spectrometry (ESI-IMS) and Raman spectroscopy was established. A database of 26 kinds of white powders, including alkaloids such as aconitine and tetrodotoxin, biological modifiers such as bradykinin, substance P and their structural analogs, biotoxins such as conotoxins, α-bungarotoxin, ricin and common white powder such as salt, flour and bovine serum albumin was constructed. Six pattern recognition algorithms of linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), k-nearest neighbor (kNN), naive bayesian (NB), classification tree (CT) and support vector machine (SVM) were used to classify the single spectrum and the fusion data based on this database. The results showed that the recognition accuracy of different methods was 76.0%–97.2%, among which the fusion recognition algorithm based on the SVM model achieved the highest recognition accuracy of 97.2%. Besides, the difference of recognition accuracy was analyzed in this study. This method could also distinguish multiple structural analogues of the two biological modifiers, and was suitable for the rapid identification of unknown white powder.

中文翻译：

---

基于数据融合算法的生物毒素和生物修饰剂快速识别研究

摘要 建立了基于电喷雾电离-离子迁移谱(ESI-IMS)和拉曼光谱的多种生物毒素和生物改性剂的分析方法。26种白色粉末的数据库，包括乌头碱、河豚毒素等生物碱，缓激肽、P物质及其结构类似物等生物改性剂，芋螺毒素、α-银环蛇毒素、蓖麻毒素等生物毒素以及食盐、面粉、构建了牛血清白蛋白。使用线性判别分析（LDA）、二次判别分析（QDA）、k-最近邻（kNN）、朴素贝叶斯（NB）、分类树（CT）和支持向量机（SVM）六种模式识别算法对样本进行分类。单光谱和基于该数据库的融合数据。结果表明，不同方法的识别准确率为76.0%~97.2%，其中基于SVM模型的融合识别算法的识别准确率最高，达到97.2%。此外，本研究还分析了识别准确率的差异。该方法还可以区分两种生物改性剂的多个结构类似物，适用于未知白色粉末的快速鉴定。