Survival analysis of patients with colorectal cancer using semi-supervised learning

Abstract

A present study introduces survival analysis for patients with Colorectal cancer. In cancer survival analysis gene selection is a significant task. The most significant invention in the clinical cancer research is to diagnose cancer more precisely dependent on the patient’s gene expression profiles. For classification of High-risk and low-risk or survival time prediction for the patient’s analytic operation, Cox proportional hazards model (COX) and accelerated failure time model (AFT) have been universally embraced. Limited number of samples and censored data are a major setback for training powerful and exact Cox classification model. Also, comparative phenotype tumors and prognoses are completely different diseases at the genotype and sub-molecular level. Subsequently, the utility of the AFT model for the survival time forecasting is restricted when such natural contrasts of the maladies have not been properly distinguished. To attempt to conquer these two fundamental issues, a novel semi-supervised learning technique has been implemented in this thesis, considering the Cox and AFT models to precisely foresee the treatment hazard and the endurance time of the patients. Furthermore, to choose the relevant genes that associate with the disease, the L1/2 regularization approach has been used in the semi-supervised learning method. Semi-supervised learning model can powerfully improve the predictive performance of Cox and AFT models in endurance examination prove in the results of simulation experiments. These methods have been effectively applied on simulated data, Diffuse large B-cell lymphoma (DLBCL_2002) microarray gene expression and clinical datasets. These methodologies were tested on new real microarray gene expression and clinical datasets of Colorectal Cancer. The upsides of the proposed semi-supervised learning technique include: Increase in the number of training samples that are available from the censored data, high capacity for distinguishing the endurance hazard classes of patients in Cox, high anticipating precision for patient’s endurance time in AFT model and robust efficiency of the proper gene selection. Semi-supervised learning model is one more applicable tool for endurance examination in clinical cancer research. The semisupervised learning method was seen to be very strong in the detection of the correct simulated genes especially when the gene expressions are independent. The analysis was performed on the real data and the results showed the semi-cox is superior compared with single cox, single AFT and semi-AFT models.