IMIPMF: Inferring miRNA-disease interactions using probabilistic matrix factorization

Jihwan Ha, Chihyun Park, Chanyoung Park, Sanghyun Park

Abstract: Highlights • A novel method (IMIPMF) for predicting miRNA-disease associations by using probabilistic matrix factorization was proposed. • IMIPMF achieved a high performance with a comparable AUC value of 0.891 by implementing five-fold cross validation. • The success of IMIPMF in predicting miRNA-disease associations can be explained by its mathematical foundations in MF and the utilization of miRNA expression data. Abstract Recently, increasing evidence have reported that microRNAs (miRNAs) play key roles in a variety of biological processes. Therefore, the identification of novel miRNA–disease associations can shed new light on disease etiology and pathogenesis. Till now, various computational methods have been proposed to predict potential miRNA–disease associations by reducing the experimental costs and time consumption. However, most existing methods are highly dependent on known miRNA–disease associations. Therefore, the prediction of new miRNAs (i.e., miRNAs without known associated diseases) and new diseases (i.e., diseases without known associated miRNAs) has become challenging. In this paper, we present IMIPMF, a novel method for predicting miRNA–disease associations using probabilistic matrix factorization (PMF), which is a machine learning technique that is widely used in recommender systems. Predicting the rating scores that a user may assign to each item in a recommender system is analogous to predicting miRNA–disease associations. By applying PMF, our model not only identifies novel miRNA-disease associations, but also overcomes the common problem of incompatibility with miRNAs without any known associated disease, which was a limitation of most previous computational methods. We demonstrated that our proposed model achieved a high performance with a reliable AUC value of 0.891 by performing 5-fold cross-validation. Overall, IMIPMF is a high-performance machine-learning-based model for predicting miRNA–disease associations, although it only considers known miRNA–disease associations and miRNA expression data.