Fig. 1

Overview of DualGCN. DualGCN takes chemical structure information of a drug and gene features of a cancer sample as inputs to the (1) drug-GCN module and (2) bio-GCN module, respectively. It outputs the response (IC50) of the given drug on the given cancer sample. (1) In the drug-GCN module, drug chemical structure data are first transformed using the previous algorithm [29]. The transformed features are considered as features of nodes (atoms). Edges between nodes represent connections between atoms of drugs. (2) The bio-GCN module is built based on PPI networks where nodes indicate cancer-related proteins (genes) and edges represent interactions between proteins. This module takes gene expression and copy number variation of cancer-related genes as inputs. Such gene features are considered as features of corresponding nodes. Embeddings from the two GCN modules are then concatenated and fed into MLP to study cancer drug response