Identification of miRNA-lncRNA Underlying Interactions Through Representation for Multiplex Heterogeneous Network

In several investigations of cancers, non-coding RNAs, especially lncRNAs (long non-coding RNAs) and miRNAs(microRNAs) have been proven that they are strongly relevant to diseases. For instance, neoplasms and non-small cell lung cancer are regulated by miRNA and lncRNA. However, it is complex that cancer could be co-regulated by multiple genes at the same time. Furthermore, different miRNAs and lncRNAs may also have interactions and regulations with others. The interactions among multiple genes still need to be interpreted. Traditional biology experiments are time-consumed and expensive. Increasing number of computational predictions of lncRNA-miRNA interactions have been seen as an alternative strategy to the biology methods for predict potential interactions. Considering that the complexity of molecular interactions, it should be more globally in identification underlying associations. We proposed a method using representation learning for attributed multiplex heterogeneous network. We conduct systematical evaluations for the model. The proposed method achieved ROC-AUC of 0.9180, PR-AUC of 0.8438, F1 scores of 0.7677. This method incorporated more biomolecular network information and provided more possibilities for discovering underlying bioinformatic associations.