摘要
目的建立基于深度神经网络的黄酮和[口山]酮高分辨质谱数据的识别分类技术。方法基于超高液相色谱-四级杆-静电场轨道阱质谱联用技术(UHPLC-Q-Orbitrap MS)对供试样品进行条件的优化和分析,使用Xcalibur 4.0软件对148个对照品的高分辨质谱数据进行提取,包括了正负离子的一、二级碎片,总计为42维,使用深度神经网络对上述数据进行二分类的建模分析,并且利用108个对照品数据作为训练集对模型进行参数调优,使其学习到区分两类化合物的能力。结果UHPLC-Q-Orbitrap MS最终采用Waters ACQUITY UPLC HSS T3(2.1 mm×100 mm,1.8μm)色谱柱进行分离,经乙腈-0.1%甲酸水梯度洗脱,流速0.2 mL·min^-1,柱温30℃,源喷雾电压正负离子分别为3.2 kV和2.8 kV;毛细管温度200℃;辅助器加热温度400℃。使用128个对照品(108个作为训练集和20个作为验证集)的高分辨质谱数据进行深度神经网络建模,并对另外20个对照品的高分辨质谱数据进行模型的测试,平均正确率达80%。结论基于UHPLC-Q-Orbitrap MS技术采集高分辨质谱数据,并使用深层前馈神经网络进行建模的分类方法,可以有效地对黄酮类和[口山]酮类化合物进行区分。
Objective To establish the recognition technology for high resolution mass spectral data of flavone and xanthone based on deep neural network.Methods The reference samples were optimized and analyzed by UHPLC-Q-Orbitrap MS.Xcalibur 4.0 software was used to extract the high resolution mass spectral data of 148 standards,including the MS and MS/MS data for both positive and negative modes,totaling 42 dimensions.The deep neural network was used to analyze the two-category modeling analysis,and the data of 108 standards were used as a training set to adjust the parameters of the model to distinguish between two types of compounds.Results The spray voltage of positive and negative modes was 3.2 kV and 2.8 kV,respectively;the capillary temperature was 200℃;the auxiliary heating temperature was 400℃.The high resolution mass spectral data of 128 standards(108 standards as the training set and 20 standards as the validation set)were used to model the deep neural network for compound classification.A correction rate of 80%in average was achieved for the test set,which was obtained from the high resolution mass spectral data of another 20 standards.Conclusion The classification model,which is trained on the high resolution mass spectral data collected by UHPLC-Q-Orbitrap MS and based on deep feed forward neural network,is reliable and may effectively distinguish flavones and xanthones.
作者
赵倩钰
王丽明
张禄
张祎
王涛
杨文志
韩立峰
ZHAO Qian-yu;WANG Li-ming;ZHANG Lu;ZHANG Yi;WANG Tao;YANG Wen-zhi;HAN Lifeng(Tianjin Key Laboratory of Traditional Chinese Medicine Chemistry and Analysis,Tianjin University of Traditional Chinese Medicine,Tianjin 301617;Shenzhen Key Laboratory of Internet of Things Key Technology,Harbin Institute of Technology(Shenzhen),Shenzhen Guangdong 518000)
出处
《中南药学》
CAS
2020年第5期748-755,共8页
Central South Pharmacy
基金
天津市自然科学基金资助项目(编号:18JCYBJC94700)。
