A clustering method with a pre-specified number ofclusters for gene selection

• Main Authors: Che-Jen Chang, 張哲仁
• Other Authors: Tzu-Tsung Wong
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/49338483517372265759

碩士 === 國立成功大學 === 資訊管理研究所 === 96 === Scientists use gene microarray (DNA microarray) data to study the differences among cells and tissues. They usually expect to find the origin causes of a specific disease. Due to the special characteristics of microarray data, high dimension and small number of samples, overfitting will generally occur, or noisy data may exist. Thus, gene selection plays an important role to solve the above problems. Though many gene selection methods have been proposed in these years, most of them are individual gene selection methods which may encounter the problems of gene collinearity and lack of consideration for combination genes. In order to solve those problems, an integration approach of individual gene ranking methods and clustering methods has been proposed to select a proper gene subset for classification. However, in such an integration approach, since the number of clusters cannot be a pre-specified parameter for clustering, its computational efficiency and stability are problematic. This research proposes an effective clustering method. After testing the clustering method on five microarray data sets, the experimental results show that it cannot only enhance the computational efficiency, but also generate stable clustering results. In addition, after applying our clustering method for gene selection, the resulting accuracies are close to the accuracies resulting from the original integration approach.