PDA-GCN: Predicting Piwi-interacting RNA-Disease Associations Based on Graph Convolution Network

Piwi-interacting RNAs (piRNAs) play a significant role in multiple complex human diseases. In order to understand disease pathogenesis at the molecular level, identifying potential associations between piRNAs and diseases accurately is very critical. The computational methods provide a cost-reducing strategy for biological wet experiments. Several computational methods have been proposed, however, most of them still adopted traditional machine learning methods. It is still a challenge for them to deal with the complex and non-linear data. Identifying associations between piRNA-disease is a link prediction task in the heterogeneous network constructed by piRNAs, diseases, and the existing verified associations between them. In this study, a computational method named PDA-GCN is proposed based on Graph Convolution Network (GCN). GCN performance well in aggregate complex information hidden in the heterogeneous network. PDA-GCN adopt 2-layer GCNs to extract features form three aspects. Experiment results show that PDA-GCN achieves superior performance comparing with state-of-the-art methods.