Identifying significant genes from microarray data

Microarray technology is a recent development in experimental molecular biology which can produce quantitative expression measurements for thousands of genes in a single, cellular mRNA sample. These many gene expression measurements form a composite profile of the sample, which can be used to differentiate samples from different classes such as tissue types or treatments. However, for the gene expression profile data obtained in a specific comparison, most likely only some of the genes will, be differentially expressed between the classes, while many other genes have similar expression levels. Selecting a list of informative differential genes from these data is important for microarray data analysis. In this paper, we describe a framework for selecting informative genes, called ranking and combination analysis (RAC), which combines various existing informative gene selection methods. We conducted experiments using three data sets and six existing feature selection methods. The results show that the RAC framework is a robust and efficient approach to identify informative gene for microarray data. The combination approach on two selecting methods almost always performed better than the less efficient individual, and in many cases, better than both. More significantly, when considering all three data sets together, the combination approach, on average, outperforms each individual feature selection method. All of these indicate that RCA might be a viable and feasible approach for the microarray gene expression analysis.