Identifying the relationship between drugs and diseases is a crucial part of the drug development process. However, discovering these relationships through experimental methods is costly and inefficient; therefore, developing efficient and accurate computational methods for predicting drug-disease associations is of great importance.In this research, a novel computational method called graph convolutional network with layer attention is proposed for predicting drug-disease associations. This model first weights the data graph using a graph attention neural network and then integrates this graph, which represents weighted drug-drug similarities, along with graphs representing known drug- disease and disease-disease associations, into a heterogeneous network. Graph convolution operations are applied to this network to learn drug and disease embedding representations. Then, the model combines representations from multiple graph convolutional layers using an attention mechanism. Finally, unseen drug-disease associations are scored based on the combined representations.eva‎luation of the proposed model shows that this method outperforms previous methods and can achieve an AUC of 95% and AUPR of 54%. A case study also demonstrates that the proposed model can effectively discover new associations. This model is a useful tool for predicting drug-disease associations. This study shows that representations from different convolutional layers can reflect different levels of proximity, and combining these representations using the attention mechanism can improve prediction performance. Furthermore, the case study shows that this model can discover new and unknown associations. The research also demonstrates the impact of changing the input graph of the network on improving the model’s accuracy.

رامين جوادي , منصوره ميرزايي

محمدرضا احمدزاده , ساره گلي فروشاني