Classification Problem Based On Gene Expression Profiles Of Cancer Research

At present, tremendous attention has been paid to the researchesof human cancer. As is known, there are many cancerous factors,including gene mutation, tumor-suppressor gene inactivation,proto-oncogene activation, and so on. Tumor-suppressor genesinactivation or proto-oncogene activation may lead to the cancerformation. In the past, many tumor-suppressor genes have beendetected, but only a few oncogenes have been determined, so findingmore cancer genes is of great significance. Cancer genes in earlystage are difficult to be discovered, so it is very important to detect thecancer genes in treatment. Early microbial genome research is limitedto a simple gene function association. With the development of thebioinformatics, gene chip analysis becomes one of the most importantmeans in cancer genes determination. Gene expression profilerepresents each gene expression instant data. Mining useful informationfrom the data and finding cancer-related genes are hot research ofcurrent bioinformatics. Based on the current colon cancer data, weestablished a model of T-statistics, signal-to-noise ratio and dynamicclustering classification, and then compare the classification results. Theclassification results can assist the research about the cancer diagnosisand treatment.The contents of this paper are as follows:In chapter1, we introduced the background, basic concepts, andmain research areas of bioinformatics and the contribution of thispaper.In chapter2, we introduced some feature selection approachesand machine-learning methods in the tumor classification problems.In chapter3, we built three classification modes of colon cancerbased on the gene expression data. It is found that the dynamic clustering method performs better than the other two methods, and theaverage prediction accuracy reaches90.62%. Our prediction resultscan assist the research on colon cancer diagnosis and treatment.