Improved Inductive Matrix Completion Method for Predicting MicroRNA-Disease Associations

摘要

Nowadays, plenty of evidence indicates that microRNAs (miRNAs) can result in various human complex diseases and may be as new biological markers to diagnose specific diseases. The reality is that biological experimental corroboration of disease-related miRNAs is time consuming and laborious. Therefore, the calculation methods for recognizing the potential relationship between the miRNA and the disease have become an increasingly significant hot topic in the world. In this paper, we exploited an improved calculation method based on inductive matrix completion to predict disease related miRNAs (I1MCMP). Firstly, we construct miRNA-disease adjacency matrix by adopting verified miRNA-disease associations. In addition, the proposed approach uses the three matrices, including the disease-miRNA association matrix, the integrated disease similarity matrix and the miRNA functional similarity matrix. Secondly, considering new diseases or new miRNAs, it is necessary to pre-process the adjacency matrix of biologically validated associations between diseases and miRNAs, so we calculate the interaction profile between miRNAs and diseases to update adjacency matrix of miRNA-disease association. Finally, inductive matrix completion algorithm is adopted to predict the probability score on the heterogeneous network between miRNAs and diseases. As a result, IIMCMP gained the AUC of 0.9016 which adopted new interaction likelihood profiles in the leave-one-out cross validation. In addition, two case studies and leave-one-out cross validation demonstrated that IIMCMP can achieve predominant and reliable performance assessment.