Neighborhood-Regularized Matrix Factorization for lncRNA–Disease Association Identification

Abstract

Long non-coding RNAs (lncRNAs) have been shown to be integral in a variety of biological processes and significantly influence the progression of several human diseases. Their involvement in disease mechanisms makes them crucial targets for research in disease biomarker identification. Understanding the intricate relationships between lncRNAs and diseases can offer valuable insights for advancing diagnostic, prognostic and therapeutic strategies. In light of this, we propose a recommendation-system-based model utilizing matrix factorization with disease neighborhood regularization to effectively infer disease-related lncRNAs (NRMFLDA). This approach leverages the power of matrix factorization techniques while incorporating disease neighborhood regularization to enhance the accuracy and reliability of lncRNA–disease association predictions. Consequently, NRMFLDA exhibits outstanding performance, achieving AUC scores of 0.9143 and 0.8993 in both leave-one-out and five-fold cross-validation, surpassing the performance of four previous models. This demonstrates its effectiveness and robustness in accurately predicting disease-related lncRNAs. We believe that NRMFLDA will not only provide innovative approaches for uncovering lncRNA–disease associations but also contribute significantly to the identification of novel biomarkers for various diseases, thereby advancing diagnostic and therapeutic strategies.