基于惩罚高斯混合模型的微阵列基因表达数据分析

Penalized Gaussian Mixture Model for the Analysis of Microarray Gene Expression Data

随着现代生物技术的发展,基于基因表达数据的肿瘤分型诊断已成为DNA微阵列的重要应用领域。提出一种基于基因表达数据的肿瘤分型诊断新方法,并在理论上给出模型解释。该方法通过对高斯混合模型加上一个L1惩罚实现了肿瘤分类和信息基因选择的有机结合,从而用较少的变量达到更高的识别率。实验结果显示,无论是在模拟数据中还是五个微阵列数据集中,提出的方法都是高效稳定的。

The diagnosis of cancer type based on gene expression profile is an important application field of DNA microarrays. In this paper, A penalized Gaussian mixture model is proposed by an Lt penalty, which makes a good conjunction with cancer diagnosis and gene selection, achieve better accuracy by less variable deal. Some theoretic analysis is given for a simple but clear explanation about the model in detail. Simulate and real data examples show that the proposed method not only have high classification accuracy but also enjoy stable performance in various cases.