基于基因灵敏度信息和二进制微粒群优化的基因选择方法

Gene selection method based on gene-to-class sensitivity information and binary particle swarm optimization

为了得到低冗余度高识别率的基因子集,提出了一种耦合基因灵敏度信息的微粒群优化基因选择方法。首先,通过单隐层神经网络从微阵列数据中提取各个基因的基因—类别灵敏度值;其次,在基因聚类基础上,利用基因灵敏度信息滤除低灵敏度的基因;最后,将基因灵敏度信息编码进二进制微粒群优化算法作进一步基因选择。在两个公开的微阵列数据集上的实验结果表明,对比其他方法,由于充分考虑各个基因灵敏度信息,因此能够选出较少基因但分类性能更高的基因子集。

To obtain more compact gene subsets resulting into high prediction accuracy on microarray data, this paper pro- posed a novel gene selection method based on particle swarm optimization （PSO） and gene-to-class sensitivity information for gene selection. To begin with, the method extracted gene-to-class sensitivity （GCS） value of every gene from microarray data by single-hidden layer feedforward neural network. Then, after all genes grouped by clustering algorithm, the method filtered out some low sensitive genes according to GCS information. Finally, the method encoded the GCS information into binary PSO （BPSO） to perform further gene selection. The experiments on two public microarray data sets verify that the proposed method can obtain better classification performance with fewer genes than other methods because of fully considering GCS information of all genes.