Cancer Classification Methods Based On Gene Expression Data

The key problem of cancer treatment lies on early prediction and full therapyeffectively and accurately. The challenge has been to distinct pathogenetical tumortypes with morphologically similar appearance and target specific therapies.Microarray technique makes it practical to quantitate the expression of thousandsof genes in parallel, and has been used in disease diagnosis and clinical test. Theapplication of gene microarray in tumor prediction and classification is beneficial toaccurately distinct tumors on molecular levels for better efficacy, as well as to probeinto pathogenesis, early prediction and molecular subtypes.Regular classification methods of gene expression data take all the genes orsamples as characteristics and do not consider that only part of genes is correlated inone biological process. This thesis presents a biclustering algorithm (HCTWC) tosearch meaningful gene signature and find natural partitions of cancer samples.The contents are listed below:(?) The meaning and development of tumor subtype classification, taking acuteleukemia (AL) and diffuse large B-cell lymphoma (DLBCL) for example.(?) Relevant knowledge of gene chips and application in tumor classification.(?) The image processing and data analysis methods of gene chips, focusing onclustering methods.(?) The types and limitations of biclustering and the design of HCTWC.(?) The realization of HCTWC based on MATLAB 6.0(?) The experiment and test of HCTWC in AL and DLBCL.By identifying relevant subsets of microarray data and focusing on them, it ispractical to discover partitions, find key genes and understand expression patternswith this algorithm, which is proved by the experiment results.