Fig. 1From: Structure-aware protein solubility prediction from sequence through graph convolutional network and predicted contact mapThe overall framework of the GraphSol model. The compiled node features (X) and contact maps (A) were utilized to construct the GCN networks, which were convoluted by two GCN blocks and layer normalizations for hidden states (M). The hidden states were converted by the self-attention layer as a fixed length of the vector (H), which were input to the MLP to make the final predictionBack to article page