Predicting Long Non-coding RNA-disease Associations using Multiple Features and Deep Learning

Various long non-coding RNAs have been shownto play crucial roles in different biological processes includingcell cycle control, transcription, translation, epigenetic regulation, splicing, differentiation, immune response and so forthin the human body. Discovering lncRNA-disease associationspromotes the awareness of human complex disease at molecular level and support the diagnosis, treatment and prevention of complex diseases. It is costly, laboratory and timeconsuming to discover and verify lncRNA-disease associationsby biological experiments. Therefore, it is crucial to develop acomputational method to predict lncRNA-disease associationsto save time and resources. In this paper, we proposed a newmethod to predict lncRNA-disease associations using multiplefeatures and deep learning. Our method uses a weighted????-nearest known neighbors algorithm as a pre-processingstep to eliminate the impact of sparsity data problem. Andit combines the linear and non-linear features extracted bysingular value decomposition and deep learning techniques,respectively, to obtain better prediction performance. Ourproposed method achieves a decisive performance with thebest AUC and AUPR values of 0.9702 and 0.8814, respectively,under LOOCV experiments. It is superior to other stateof-the-art SDLDA and NCPLDA methods in both AUC andAUPR evaluation metrics. It could be considered as a powerfultool to predict lncRNA-disease associations.