Studies On Gene Signature Based On Microarray Data And Applications In Breast Cancer Data Analysis

As a high-throughput technology, microarray has been widely adopted in the research of human diseases. In microarray experiment, the expression of thousands of genes or even all genes in genome can be observed simutaneously, and the high demension data need special statistic analysis. Many classification algorithms in machine learning have been utilized to analyze microarray data, such as gene signature selection. Gene signature selected on microarray data can support personal prognosis, divide potential subtypes of diseases and help to choose suitable treatment according to the patient gene expression profiling pattern. In this thesis, we developed a research flow of extract gene signature from breast cancer microarray data, it comprises microarray data quality control, choosing differentially expressed genes, predicting sample's state with certain classification algorithm and external validation. This work flow can extend to other microarray data analysis for disease gene signature discovery.Due to microarray experiment complexity protocol, various platform, sample' s difference, dissimilar experiment designs and multiplicate data analysis strategies, independent researchers who focus on the same question often had different results and got different gene signatures. In this thesis, we proposed an algorithm to explore the relation of different gene signatures. In the algorithm, we designed a matrix of gene expression profiling correlation coefficient of two different gene signatures, and found the relation of the two signatures through connectivity path searching on the matrix.