Predicting lncRNA-Disease Associations through Attention Graph Convolutional Autoencode
Xiaoqian Li; Shanwen Zhang; Shenao Yuan
This thesis takes lncRNA, disease as the research object, and starts from the aspects of heterogeneous network construction, feature information extraction, and association prediction model establishment, respectively, and proposes an improved graph self-encoder prediction model (AGCELDA for short) for predicting disease-related lncRNAs, in response to the problem of incomplete modelling of association relationships in the existing lncRNA-disease association prediction model. The model Firstly, a heterogeneous graph is constructed using similar information of lncRNAs, diseases and known lncRNA-disease associations; subsequently, a graph encoder embedded with an attention mechanism is used to model the lncRNA-disease association relationship to obtain high-quality, low-dimensional representation vectors; and finally, the graph decoder is used to reconstruct the association relationship between lncRNAs and diseases for potential association prediction. Using this method can better capture the association relationship between nodes in the heterogeneous graph, thus effectively improving the performance of lncRNA-disease association prediction. In the experiments, the model was subjected to five-fold cross-validation and compared with other methods, and the experiments proved that the AUC accuracy of the AGCELDA method was relatively high. Finally, a case study was also conducted to demonstrate the capability of AGCELDA in identifying candidate lncRNAs that may be associated with diseases.