Fig. 4
From: GPDRP: a multimodal framework for drug response prediction with graph transformer
Illustration of the predictive analysis workflow of GPDRP. A Drug molecular graph construction. The structure information of drugs was collected from PubChem and we represented drugs as molecular graphs using RDKit. B Gene pathway activity scores calculation. For the cancer cell lines obtained from CCLE, we computed pathway activity scores for canonical pathways using GSVA. C Two subnetworks for learning drug features and cell line features respectively. GPDRP took molecular graphs of drugs and gene pathway activity scores of cell lines as inputs to the drug subnetworks and cell line subnetworks, respectively. The two representations are then concatenated and put through two FC layers to predict the response. D Results and downstream analysis of this work. Including performance comparison, prediction of unknown drug-cell line response and predictions in LNCaP xenografts