A Constrained Probabilistic Matrix Decomposition Method for Predicting miRNA-disease Associations

Background: MicroRNA is a type of non-coding RNA molecule whose length is about 22 nucleotides. The growing evidence shows that microRNA makes critical regulations in the development of complex diseases, such as cancers, and cardiovascular diseases. Predicting potential microRNA-disease associations can provide a new perspective to achieve a better scheme of disease diagnosis and prognosis. However, there is a challenge to predict some potential essential microRNAs only with few known associations. Objective: In this paper, we propose a novel method, named as a constrained strategy for predicting microRNA-disease associations called CPMDA, which can predict some potential essential microRNAs only with few known associations. Methods: We firstly construct a disease similarity network and microRNA similarity network to preprocess the microRNAs with none available associations. Then, we apply probabilistic factorization to obtain two feature matrices of microRNA and disease. Meanwhile, we formulate a similarity feature matrix as constraints in the factorization process. Finally, we utilize obtained feature matrixes to identify potential associations for all diseases. Result and Conclusion: The results indicate that CPMDA is superior over other methods in predicting potential microRNA-disease associations. Moreover, the evaluation shows that CPMDA has a strong effect on microRNAs with few known associations. In case studies, CPMDA also demonstrated the effectiveness to infer unknown microRNA-disease associations for those novel diseases and microRNAs.