Using a deep learning approach to identifysurvival subtypes of esophageal squamous cell carcinomapatients

Background: Esophageal squamous cell carcinoma (ESCC),with a poor prognosis, is one of the most lethal types of cancer.In the present study, we used state-of-the-art machine learningand deep learning techniques to stratify the survival of esophagealsquamous cell carcinoma Patients by lncRNA profiles.Materials and Methods: Expression profiles, demographicand clinical variables of ۶۰ patients were downloaded fromGene Expression Omnibus (GEO) dataset. We used a deeplearning (DL)– autoencoder model to extract the LncRNAs andthe univariate Cox proportional hazards model to select significantfeatures obtained from DL method (P<۰.۰۵). Then hierarchicalclustering used to identify High and low-risk groupsbased on the obtained significant features.Results: A thousand features from ۸,۹۰۰ LncRNAs extractedby using the Autoencoder framework in the bottleneck of thenetwork. Using the univariate Cox-PH model, ۴۳ features weresignificantly identified related to time-to-death from ESCC,which were used as the inputs for the clustering of patients.Then two high-risk and low-risk groups were identified based on the hierarchical clustering. Using the Kaplan Meier's curve,the patients with higher than median survival was considered asthe low-risk survival patients, and the other group was consideredas the high-risk group (P-value of log-rank test = ۰.۰۲۲).Conclusion: This study identified prominent lncRNAs thathaving a role in determining high-risk patients for death ofESCC. Our findings are intriguing and represent an importantstep forward. Further research is needed to fully validate thepotential of these novel lncRNAs and to establish their clinicalutility.