Abstract
Background: Colon cancer is the second leading cause of human death worldwide. Recently, there has been an increasing amount of studies investigating colon cancer stem cells (CCSCs), which encompass characteristics of self-renewal, mature tumor cells generation through differentiation, drug resistance and metastasis. Patients with genetic signatures of CSCs were shown to have worse survival rate and higher chance of cancer relapse. It is hence of interest to identify novel prognostic markers for colon cancer relapse via the genetic signatures of CSC markers. Result: Based on 12 primary integrated microarray datasets, we divided colon cancer patients into marker- and marker+ subgroups according to 8 well-known CSC markers. Then, we constructed their corresponding protein-protein interaction networks (PPINs) for both marker- and marker+ subgroups. We then proposed and calculated prognostic proteins relevance values (PPRV) and obtained a total of 12 significant proteins high in PPRVs for colon cancer patients. In addition, we applied a deep learning algorithm (a deep neural network, DNN) and support vector machine (SVM) algorithm to predict the 5-year relapse rate of colon cancer patients. Our results showed that the prediction accuracy of 12 of our identified markers (AUC = 0.811; Accuracy = 77.3) is much better than the 8 well-known CSC markers (AUC = 0.738; Accuracy = 72.7) via DNN; and the prediction accuracy of 12 of our identified markers (AUC = 0.816; Accuracy = 79.0) is again much better than the 8 well-known CSC markers (AUC = 0.766; Accuracy = 76.9) via SVM. Conclusion: According to our constructed PPINs and calculated PPRVs, we were able to identify 12 novel prognostic markers for colon CSC markers. This study integrates systems biology and deep learning methodologies that allows for the identification of novel prognostic markers via existing CSC markers, and accurate prediction of colon cancer 5-year relapse. We believe that such a systemic approach to uncover novel cancer prognostic markers and to predict cancer relapse provides the foundation for the design of better therapeutic strategy for cancer treatment in the future. Keywords: Cancer stem cells, colon cancer, prognostic markers, deep learning, SVM, systems biology.