Fig. 2
a Encode each node feature via BiLSTM. b Aggregate neighbors to obtain drug and protein embedding, predict label via a two-layers neural network, finally optimizes the model via a cross-entropy loss
Fig. 2 | BMC Bioinformatics