基于异质图注意力网络的miRNA与疾病关联预测算法

Lots of experiments have shown that the abnormal expression of microRNA(miRNA) is related to the evolution and progression of human complex diseases. Identifying associations between miRNAs and diseases is beneficial to promote the development of clinical medicine. However, traditional experimental methods are often time-consuming and inefficient, so there is an urgent need for efficient computational methods to predict the potential associations between miRNAs and diseases. In this paper, we propose an end-to-end computational model based on heterogeneous graph attention network to predict the associations between miRNAs and diseases. This model captures the structure and attribute information of heterogeneous neighbors via the multi-head attention mechanism, and fuses them with the attribute information of the central vertex to generate more representative feature embeddings of miRNAs and diseases, and then predicts the potential associations between miRNAs and diseases through a fully connected layer. The 5-fold cross-validation results show that our model achieves 93.52% and 94.82% AUC values based on HMDD v2.0 and HMDD v3.0 datasets, respectively. In addition, the case study on esophageal neoplasms shows that 48 of the top 50 miRNAs predicted by our model are confirmed. The above experimental results indicate that our model can be used as a reliable tool to predict candidate disease-related miRNAs.