基于小波变换的代谢组色谱指纹图谱的判别分析  

Discriminant Analysis Based on Wavelet Transform in Metabolomics Liquid Chromatography Data

作　　者：孙凤宇[1] 李贞子[1] 侯艳[1] 李康[1] 

机构地区：[1]哈尔滨医科大学卫生统计学教研室,150081

出　　处：《中国卫生统计》2013年第2期206-208,共3页Chinese Journal of Health Statistics

基　　金：国家自然科学基金项目(81172767);哈尔滨医科大学伍连德青年基金(WLD-QN1105)

摘　　要：目的探索小波变换在代谢组学超高效液相色谱信号分析中的作用。方法将卵巢癌和卵巢囊肿的色谱数据使用不同的连续小波基函数和不同尺度变换,随机抽取一定数量的训练数据,采用随机森林(RF)方法进行特征筛选建立模型,最后对测试数据集进行判别分析,计算ROC曲线下面积AUC值。结果经过小波变换预处理后的数据建模分类效果明显优于原始数据,其中墨西哥帽(mexh)小波变换分类效果最好。结论基于小波变换的代谢组特征提取方法能够明显提高模型的判别能力,具有研究价值。Objective To explore the influence of the wavelet transformation on the signal decomposition of ultra performance liquid chromatography. Methods We use different continuous wavelet base functions with scales changing for the chromatographic data from ovarian cancer and benign tumors, and random forest model is used to conduct fea- ture selection for training samples and discriminate analysis for test sam- pies,then the area under the ROC curve is calculated. Results The clas- sification performance with the wavelet transformation data is better than with the original data, where the performance with the Mexican hat wavelet is better than other wavelet base functions. Conclusion The feature se- lection method based on the continuous wavelet transformation and the dis- crimination model can significantly improve the classification ability and is worthy of further study.

关 键 词：代谢组 小波变换 色谱 随机森林 

分 类 号：R312[医药卫生—基础医学]