LncRNA Expression Profile-Based Matrix Factorization for Predicting lncRNA- Disease Association

Long non-coding RNAs (lncRNAs) play significant roles in multiple biological processes and contribute to the progression and development of various human diseases. Therefore, it is necessary to decipher novel lncRNA-disease associations from the perspective of biomarker detection. Numerous computational models have been designed to identify lncRNA-disease associations using machine learning. However, many of these models fail to effectively incorporate heterogeneous biological datasets, which can lead to reduced model accuracy and performance. In this study, we propose a novel lncRNA expression profile-based matrix factorization method that applies lncRNA expression profiles to identify lncRNA-disease associations (EMFLDA). Matrix factorization is a machine learning method that exhibits excellent performance not only in recommender systems, but also in various scientific areas. We also applied lncRNA expression profiles as weights for the proposed model, which allowed for the integration of heterogeneous information and thereby improved performance. As a result, EMFLDA outperformed the four previous models in terms of AUC scores, achieving scores of 0.9042 and 0.8841 based on leave-one-out cross-validation and five-fold cross-validation, respectively. Thus, the proposed model, EMFLDA, not only serves as an effective tool for identifying disease-related lncRNAs, but also plays a pivotal role in extracting disease biomarkers.