A New Gene Selection Method Based on PCA for Molecular Classification

Microarray expression experiments generating thousands of gene expression measurements simultaneously provide information for tissue and cell samples, which are useful for disease diagnosis. These experiments primarily either monitor each gene multiple times under different conditions or alternatively evaluate each gene in a single environment but in different types of tissues. In general, microarray data are huge and difficult to analyze. In order to extract information from gene expression measurements, various methods have been employed to analyze this data such as SVM, clustering methods, self-organizing maps, and weighted correlation method. Support vector machines have been shown to perform very well in many areas of biological data analysis, in particular microarray expression data analysis. We present a new gene selection method for microarray data analysis. This method removes noisy data using principal component analysis, and selects genes with high contribution to constitute principal components. Selected genes have discriminative power to distinguish classes. When we used the presented method with SVM, we were able to analyze microarray data more correctly than previously known methods for molecular classification.