Determination of Colorectal Cancer and Lung Cancer Related LncRNAs based on Deep Autoencoder and Deep Neural Network

Until recently, non-coding RNAs were considered junk RNA and were always ignored, but studies have revealed that many non-coding RNAs such as miRNA, lncRNA, and circRNAs play important roles in biological processes. A subclass of non-coding RNAs with transcripts longer than 200 nucleotides, called lncRNAs, play important roles in many cellular processes such as gene regulation. For this reason, since wet experimental studies to identify disease-related lncRNA are time-consuming, computational methods are used. Many researchers have applied similarity-based and machine learning-based computational methods and achieved very successful results. Due to its high success rate, the deep learning technique is applied to many fields today. In this study, we used the Deep Autoencoder and Deep Neural Network method to predict disease related lncRNAs. As input data of Deep Autoencoder, the concatenated feature vector obtained from integrated disease similarity and integrated lncRNA similarity was used. To train the deep neural network for predicting relationships between lncRNAs and diseases, the features extracted from the autoencoder’s output were utilized. The prediction performance of our method was evaluated with the commonly used 5-fold cross validation and an AUC value of 0.9575 was obtained. It can be seen that the method we proposed is more successful than other compared methods. Additionally, case studies on colorectal cancer and lung cancer were conducted and confirmed with the literature. As a result, the Deep Autoencoder and Deep Neural Network method can be used reliably to identify candidate disease-related lncRNAs.