Background: MiRNAs play an important role in complex human diseases, and the identification of disease and miRNA associations can accelerate drug development, individualized diagnosis, and treatment of diseases. From a miRNA perspective, the molecular mechanisms of many complex diseases, such as metabolic disease, are not fully understood. However, experimentally exploring the relationship between miRNA and disease is expensive and time-consuming. The purpose of this study was to examine whether the computational model could infer related miRNAs for metabolic disease.
Methods: miRNA and disease can construct a bipartite graph according to their associations. Here, we propose a novel graph-based two-way diffusion model to reconstruct the bipartite graph and then transform the prediction task into a matrix to complete the problem. The underlying assumption made by our model is that miRNA or diseases that are highly similar to each other tend to have similar correlations and vice versa. We calculate the miRNA similarities for all the miRNAs by use MiRNA functional similarity and miRNA Gaussian similarity. We also calculate the semantic similarity and Gaussian similarity for all the diseases. Access to a miRNA expression, our model can infer its correlation with each metabolic disease.
Results: To assess the predictive model, we used the HMDD V3.2 database, which included 35,547 miRNA-disease association entries consisting of 1,206 miRNA and 893 diseases identified from the literature. The results show that the prediction model can achieve AUCs of 0.916 based on 5-fold cross-validation. We took type 2 diabetes mellitus as a target for miRNA prediction, 9 out of the top 10 predicted potential miRNA were confirmed by literature (e.g., hsa-mir-205).
Conclusions: We can use the model to select the most likely miRNA targets that specific metabolic diseases would associate with. It is anticipated that our model would become a practical tool for the clinical metabolic diagnosis and lab experiments.
P. Hu: None. Z. Kuang: None. T. He: None. Y. Huang: None. Z. Tang: None. S. Li: None. J. Mei: None.
National Natural Science Foundation of China (61702424)