肿瘤微阵列数据的小波模极大值特征提取

本文提出针对肿瘤微阵列数据的小波模极大值特征提取方法。首先求两类数据的Bhattacharyya 距离分布，初 步提取特征基因；接着进行小波分解，在频域上用小波分解高频系数检测基因突变点，低频系数逼近表征原始信号特征；然后 通过理论分析和构建SVM 分类器，经过多次实验选取小波基和尺度，提取特征基因。将该算法应用于数据集(1999 年Golub 所 用ALL 和AML)，从中提取了5 个基因，分类测试准确率可达94.12%。可见该算法具有较高的可行性与有效性，能为肿瘤间差 异基因研究提供一定参考。

Feature Extraction Base onWavelet Modulus Maxima for Microarray Data

A new method of microarray data to extract features based on wavelet modulus maxima is proposed in this paper. First of all, the Bhattacharyya distance distributions of two classes are derived, preliminarily extracting feature genes. Then wavelet decomposition is adopted to detect the gene mutation of high frequency coefficient, and to approximate the original signal characterization based on low frequency. Finally the features are extracted by theoretical analysis and SVM classification, which selects the wavelet basis and scale based on multiple experiments. The proposed method is applied on the data set (1999 Golub used in ALL and AML). Five feature genes are extracted, whose classification test accuracy rate can reach 94.12%. It can be seen that the algorithm has high feasibility and effectiveness, and can provide some reference for the study of the differentially expressed genes between tumors.