乳腺癌基因芯片数据分析

目的　以乳腺癌病人的表达谱芯片数据为基础 ,探寻乳腺癌复发的相关基因。方法　对标化芯片数据进行缺失值处理后 ,分别用单因素COX回归模型和综合了聚类及多因素COX回归的综合法来筛选兴趣基因 ,然后通过兴趣基因对病人做样品聚类 ,以灵敏度、特异度、约登指数和Kaplan Meier法评价分类效果 ,最后结合文献和蛋白质数据库探寻乳腺癌复发的相关基因。结果　综合法筛出的 30个P <0 .0 1的基因对乳腺癌病人复发状况的预测效果最佳 ;单因素COX回归筛出的 1 0 2个P <0 .0 1的基因的预测效果较差 ;单因素COX回归筛出的 1 5个P <0 .0 0 1的基因预测效果最差。结论　综合法筛选得到的 30个基因可用来评价病人的预后状况 ,为进一步的生物学研究提供待选基因。

Microarray Data Analysis for Breast Cancer

Purpose To study the relapse-related genes based on gene expression profiles from breast cancer patients with different clinical outcomes. Methods Firstly,univariate COX regression model was used to analysis the microarray data to select the potential genes.Secondly,an integrated method composed of cluster and multivariate COX analysis was also carried out.Then K-means cluster method was applied to classify the relapse situation of patients.Sensitivity,specificity,Youden's index and Kaplan-Meier analysis were used to evaluate the genes selected by different methods. Results The 30 genes(P<0.01)selected by the integrated method performed best when predicting the relapse situation of patients.The 102 genes(P<0.01) by univariate COX analysis performed well while the 15 genes(P<0.001)by univariate COX analysis performed worst. Conclusions The 30 genes selected by the integrated method,especially the genes which were also picked out by the other two methods,are worthy of further experiments to assess the results of microarray.