MicroRNAs (miRNAs) are small single-stranded noncoding RNAs that have shown to play a critical role in regulating gene expression. In past decades, cumulative experimental studies have verified that miRNAs are implicated in many complex human diseases and might be potential biomarkers for various types of diseases. With the increase of miRNA-related data and the development of analysis methodologies, some computational methods have been developed for predicting miRNA-disease associations, which are more economical and time-saving than traditional biological experimental approaches. In this study, a novel computational model, deep belief network (DBN)-based matrix factorization (DBN-MF), is proposed for miRNA-disease association prediction. First, the raw interaction features of miRNAs and diseases were obtained from the miRNA-disease adjacent matrix. Second, 2 DBNs were used for unsupervised learning of the features of miRNAs and diseases, respectively, based on the raw interaction features. Finally, a classifier consisting of 2 DBNs and a cosine score function was trained with the initial weights of DBN from the last step. During the training, the miRNA-disease adjacent matrix was factorized into 2 feature matrices for the representation of miRNAs and diseases, and the final prediction label was obtained according to the feature matrices. The experimental results show that the proposed model outperforms the state-of-the-art approaches in miRNA-disease association prediction based on the 10-fold cross-validation. Besides, the effectiveness of our model was further demonstrated by case studies.

MicroRNA（miRNA）是小的单链非编码RNA，已显示在调节基因表达中起关键作用。在过去的几十年中，累积的实验研究已经证实，miRNA与许多复杂的人类疾病有关，并且可能是各种疾病的潜在生物标记。随着与miRNA相关的数据的增加和分析方法的发展，已经开发了一些用于预测miRNA-疾病关联的计算方法，比传统的生物学实验方法更经济，更省时。在这项研究中，提出了一种新的计算模型，即基于深度信念网络（DBN）的矩阵分解（DBN-MF），用于miRNA-疾病关联预测。第一，从miRNA疾病相邻基质中获得了miRNA与疾病的原始相互作用特征。其次，基于原始的交互特征，分别使用2个DBN进行miRNA和疾病特征的无监督学习。最后，使用从最后一步开始的DBN初始权重训练由2个DBN和一个余弦得分函数组成的分类器。在训练过程中，将miRNA疾病相邻矩阵分解为2个用于表示miRNA和疾病的特征矩阵，并根据特征矩阵获得最终的预测标签。实验结果表明，该模型在基于10倍交叉验证的miRNA疾病关联预测中优于最新方法。除了，