基于随机森林算法的原位质谱快速鉴别肺癌的方法研究  

Mass Spectrometric Discrimination of Human Lung Tumors under Ambient Conditions Based on Random Forest Algorithm

作　　者：欧阳永中 曾玉庭 郭伟清 邓金连 魏益平[3] OUYANG Yong-Zhong;ZENG Yu-Ting;GUO Wei-Qing;DENG Jin-Lian;WEI Yi-Ping(School of Environmental and Chemical Engineering,Foshan University,Foshan 528000,China;School of Food Science and Engineering,Foshan University,Foshan 528000,China;Department of Cardiothoracic Surgery,Second Affiliated Hospital of Nanchang University,Nanchang 330006,China)

机构地区：[1]佛山科学技术学院环境与化学工程学院,佛山528000 [2]佛山科学技术学院食品科学与工程学院,佛山528000 [3]南昌大学附属第二医院胸心外科,南昌330006

出　　处：《分析化学》2020年第8期1012-1017,I0001-I0005,共11页Chinese Journal of Analytical Chemistry

基　　金：国家自然科学基金项目(No.21405013)资助。

摘　　要：随机森林(Random forest,RF)算法是一种基于决策树的机器学习算法,具有良好的分类与变量筛选性能,因而在生物医学高维数据分析中应用广泛。本研究开发了一种基于RF算法的原位质谱快速鉴别肺癌的模型和方法,通过构建液体辅助表面解吸常压化学电离质谱技术平台(DAPCI-MS),结合RF算法,在常温常压条件下,直接实现对未处理人体肺鳞癌组织切片的准确鉴别与区分,并获取肺癌区别于正常组织的生物特征标记物。研究表明,当决策树数目n tree=100时,对人体肺鳞癌组织与邻近正常组织的区分准确率可达到100%。与其它分类方法相比,本模型具有稳健性高、分类效果好、泛化能力强等特点,为实现复杂基质的人体肺癌组织与相邻正常组织的区分提供了一种快速、准确和可靠的分类模型。Random forest algorithm(RF)is a machine learning algorithm based on decision trees.Due to the good performance of classification and variables selection,it has been widely used in biomedical high-dimensional data analysis.In order to fast and accurately distinguish human lung cancer from adjacent normal tissues,a model for direct ambient mass spectrometric analysis of lung cancer tissue sections based on random forest algorithm was developed.The purpose of this study was to establish a liquid assisted surface desorption atmospheric pressure chemical ionization mass spectrometry(DAPCI-MS)platform,combined with the random forest algorithm,to directly identify and differentiate the untreated human lung squamous cell carcinoma tissue sections under normal temperature and pressure,as well as obtaining the biomarkers of lung cancer for differentiation from normal tissue.The results showed that when the number of decision trees n tree=100,the accuracy of distinguishing human lung squamous cell carcinoma from adjacent normal tissues reached 100%.Compared with other methods,this model had higher robustness,better classification effect and stronger generalization ability.This study provided a more accurate and reliable classification model for rapid differentiation of human lung cancer tissues from adjacent normal tissues in complex matrix.

关 键 词：随机森林算法 表面解吸常压化学电离质谱技术 肺癌组织切片 特征生物标记物 

分 类 号：R734.2[医药卫生—肿瘤] O657.63[医药卫生—临床医学]