With critical roles in regulating gene expression, miRNAs are strongly implicated in the pathophysiology of many complex diseases. Experimental methods to determine disease related miRNAs are time consuming and costly. Computationally predicting miRNA-disease associations has potential applications in finding miRNA therapeutic pathways and in understanding the role of miRNAs in disease-disease relationships. In this study, we propose the MiRNA-disease Association Prediction (MAP) method, an in-silico method to predict and prioritize miRNA-disease associations. The MAP method applies a network diffusion approach, starting from the known disease genes in a heterogenous network constructed from miRNA-gene associations, protein-protein interactions, and gene-disease associations. Validation using experimental data on miRNA-disease associations demonstrated superior performance to two current state-of-the-art methods, with areas under the ROC curve all over 0.8 for four types of cancer. MAP is successfully applied to predict differential miRNA expression in four cancer types. Most strikingly, disease-disease relationships in terms of shared miRNAs revealed hidden disease subtyping comparable to that of previous work on shared genes between diseases, with applications for multi-omics characterization of disease relationships.

在调节基因表达中起关键作用，miRNA与许多复杂疾病的病理生理密切相关。确定疾病相关miRNA的实验方法既费时又费钱。以计算方式预测miRNA-疾病的关联在发现miRNA治疗途径和了解miRNA在疾病-疾病关系中的作用方面具有潜在的应用。在这项研究中，我们提出了一种miRNA-疾病关联预测（MAP）方法，一种计算机模拟预测和区分miRNA疾病关联的方法。MAP方法应用网络扩散方法，从异源网络中的已知疾病基因开始，该异源网络由miRNA-基因关联，蛋白质-蛋白质相互作用和基因-疾病关联构成。使用关于miRNA疾病关联的实验数据进行的验证表明，与两种当前的最新技术相比，其优越的性能，其中四种类型的癌症的ROC曲线下面积均超过0.8。MAP已成功应用于预测四种癌症类型中差异的miRNA表达。最令人惊讶的是，就共享miRNA而言，疾病-疾病关系揭示了与先前关于疾病之间共享基因的研究相当的隐藏疾病亚型，并应用了多组学方法表征疾病关系。