Fig. 1
From: Fusing graph transformer with multi-aggregate GCN for enhanced drug–disease associations prediction
The overall architecture of the proposed WMAGT. WMAGT involves three main steps. First, drug and disease similarity networks are jointly encoded using GCN and graph transformer for representation projection. In the second step, matrix operations project drug and disease representations in the network, generating new information. Lastly, the domain information from the first step and interactive information from the second step are utilized in the NCF module, and multiple loss functions along with MLP are employed to comprehensively model the drug–disease relationship