Abstract
Novel biological technology based on microarray experiments has been extensively applied by various fields of researches due to the utility of microarray capable of investigating tens thousands of genes simultaneously. However, gene analysis lacks an organized process to effectively discover crucial genes from image acquisition. Therefore, one computation model was developed to obtain the most differentially expressed genes and the most discriminative genes. We started gene discovery from computing gene expression levels of 44 microarray images by the Otsu thresholding method. Next, the ratios of Cy3 and Cy5 fluorescence intensities of tumor and normal samples respectively of each spot were normalized based on piecewise linear regression method. Finally, thresholding strategies and feature selection methods were utilized to acquire significant genes, accompanied by clustering algorithms to verify the suitability of the selected genes. Ninety percent of correlation coefficients between our computing data sets of gene expression levels and the ones generated by commercial software (ArrayPro Analyzer) were larger than 0.5. Ninety percent of correlation coefficients between the data sets of our normalized log-ratios and those normalized by LOWESS regression method had values larger than 0.9. These results suggest that our computational model is practical and efficient. Results in this thesis provide informative materials of computational and quantitative examination for microarray-related research and facilitate clinical diagnosis and gene analysis.