论文部分内容阅读
基于肿瘤基因表达谱,提出了单隐层BP网络灵敏度分析法以有效选取肿瘤亚型分类特征基因。首先以Bhattacharyya距离为尺度滤除分类无关基因;然后从“功能基因组合”的角度出发,依据输入特征对BP网络输出的灵敏度生成候选特征基因子集;最后以BP网络作为样本分类器考察候选特征基因子集对肿瘤样本的亚型识别能力,得到具有最佳分类能力的基因子集作为肿瘤亚型分类特征基因。以小圆蓝细胞瘤基因表达谱数据集为例进行实验,结果表明了该方法的可行性和有效性。
Based on the tumor gene expression profile, a single hidden layer BP network sensitivity analysis method was proposed to effectively select the tumor subtypes. Firstly, Bhattacharyya distance is used as a criterion to filter out the irrelevant genes. Then from the perspective of “functional gene combination ”, a subset of candidate feature genes is generated based on the sensitivity of the input features to the BP network output. Finally, BP network is used as a sample classifier Candidate characteristic gene subsets of tumor samples subtype recognition ability, get the best classification ability of the gene subset as a tumor subtype classification features genes. The small round blue cell tumor gene expression data set as an example experiment, the results show that the method is feasible and effective.