Dimensionality Reduction Of Breast Cancer Patients’ Clinical Data And Survival Prediction Analysis

With the development of gene technology, we can easily get the mi RNA expression data of cancer patients. The mi RNA plays an important role in cancer metastasis in recent papers. It attracts our attention that we can analyze the patient’s condition from the miRNA expression. In this thesis, we study the dimensionality reduction of high-dimensional miRNA expression data and survival predicti on analysis of clinical cancer data.Because the patient’s clinical data contains a large number of censored data, we cannot ignore censeored data to receive the survival distribution fun ction. Therefore, we use two methods, product limit estimation method and maximum likelihood estimation method, to get the survival distribution function of breast cancer patients. We prove that maximum likelihood estimation method has a unique solution under the certain conditions. Meanwhile, we propose an interpolation method with the same distribution as the original data to implement interpolation. Then we analyze the error of the interpolation method with this given distribution.High demension is another typical feature of cancer patients’ clinical data. Accordingly, high-demensional data processing is proposed to implement dimensionality reduction, which can be categorized as feature selection method and feature extraction method. We compare the minimum mesh clustering, principal component analysis and Isomap methods for this purpose. We find that the shortest path of Isomap method indicates the development path of cancer. Then we modify minimum mesh clustering method by updating decentralization. Finally, we establish classifiers for cancer stagings based on mi RNA expression data after dimension reduction. We compare the support vector machine classifier with the decision tree classifier.Finally, we identify some mi RNA associated with cancer staging, by using the method of the Kruskal-Wallis test of mi RNA expression data and clinical data. We use the stepwise multiple regression method to establish the regression relationship between mi RNA expression and patient’s survival time. The mi RNAs(hsa-mir-548 t, hsa-mir-190, hsa-mir-200 b, etc) selected by stepwise multiple regression model has proved to be associated with cancer in literature.