基于人工神经网络集成的微阵列数据分类

Microarray data classification based on artificial neural network ensembles

基因数量远多于样本数量是基因表达微阵列数据进行疾病诊断所面临的主要挑战,为此提出了采用人工神经网络集成的组织分类方法.该方法使用Wilcoxon测试选择用于与分类相关性较高的重要基因,通过凸伪数据法产生神经网络集成中各个体的训练集,用简单平均法集成网络个体的测试结果.采用实际的微阵列实验数据集分别进行独立测试和交叉验证测试,结果表明,该方法性能优于单个神经网络、最近邻法和决策树.受试者特征曲线测试表明,神经网络集成性能优于单个神经网络.

Disease diagnostics based on gene expression microarray data presents major challenges due to the number of genes far exceeding the number of samples. A tissue classification method using artificial neural network ensembles was proposed. In this method, significant genes highly correlated with classification were selected by Wilcoxon test. Then training data sets for each member of neural network ensembles were generated by convex pseudo-data algorithm. The predictions of those individual neural networks were combined by simple average method. The results of independent tests and cross validation tests using real micorarray experimental data sets show that this classification method outperformed than single neural networks, 1-nearest-neighbor classifiers and decision trees. Receiver operator characteristics curve tests demonstrate that the performance of neural network ensembles is better than that of single neural networks.