A general framework for functionally analyzing microarray data

With the growing achievements of the Human Genome Project, huge sets of biological data are being generated for the integrated study of understanding biological processes responsible for different phenomena. As the emergent microarray technology starts to produce large amounts of diverse data, the need for reliable data analysis tools arises. Further, integrating different sets of data require special, application-dependent considerations. In this dissertation, we present a general signal processing dimension reduction framework that significantly improves finding strong feature (gene) sets based on similar and dissimilar patterns through the joint analysis of gene expression and DNA copy number microarray data.; A microarray experiment can be broadly divided into three stages: (i) array fabrication, (ii) probe preparation and hybridization, and (iii) data collection, normalization and analysis. The current work concentrates on the computer-assisted data analysis tools in the last stage for pointed biological research questions regarding gene selection. In this dissertation, analysis tools are presented for the joint study of gene expression levels and DNA copy number variations, but our methods are tractable for other input data types. We specifically focus on the application of analyzing breast cancer cell lines and tumors. As we approach the problem of locating biological markers for disease, we need to keep in mind the fundamental issue of a limited number of samples (tissues) with many variables (genes). Dimension reduction models are employed to overcome these limitations. In addition, low level analysis is paramount for obtaining useful data for analysis and we explore data normalization considerations. We also uncover statistically independent underlying sources of variation. Ultimately, we propose a general framework to specifically address all these concerns. In the presentation of our findings, we provide a statistical assessment when appropriate. In addition, we have developed visualization tools to assist biologists utilize and understand the outputs from our analysis methods.