Development of an approach to predict the recurrence of cancer

Abstract

There has been abundant research to predict the recurrence of cancer. Several informative factors, such as microRNA, gene expression, clinical data, and images of tissue and cells, have been used to predict cancer recurrence. Most neoplasm, however, originates from a single cell, and tumor progression results from the genetically unstable cell by acquiring a genetic variant within the original clone. Thus, the importance of understanding the progression of the tumor is emphasized. In addition, the acquired genetic instability results in very individual biological outcomes for advanced human malignant tumors. In this study, we propose an approach to infer the clonal expansion of DNA to identify tumor progression and find that the mutation of the genes that typically mutates in the early stage of cancer is a recurrence risk factor. Moreover, this study provides insight on new personalized therapies on the basis of patients' individual mutations by identifying the genetic variants driving cancer recurrence. Our study can be divided into three parts. The first is an approach for obtaining the cellular prevalence (CP) value to find candidate genes. The second is classifying patients with a high risk of cancer recurrence, and the third is constructing machine-learning models to predict the recurrence of cancer in patients. We analyzed SNPs from tumor tissue in five types of cancer patients in the TCGA, including 784 samples of WXS data. The patients in the group with a high risk of cancer recurrence show a poor prognosis in KIRC($P <$ 0.05) and both prognosis and survival in the PAAD($P <$ 0.005). Additionally, when using the feature HIGH5 for predicting cancer recurrence in patients, accuracy and AUC are higher than without using HIGH5. In this study, we can classify patients with a high probability of recurrence. We manually constructed the approaches to infer tumor progression. Also, we propose the features that can be used to predict recurrence by machine learning techniques.