Fig. 1

Scheme for utilizing atom-wise featurization and topological information on compounds for the identification of pan-kinase family inhibitors (PKFIs) using graph convolutional network (GCN) models. A Schematic of the research framework. B 195,802 test datasets of 60,122 chemicals and 384 kinases were collected from the ChEMBL database and kinase profiling. C Eight families in four kinase groups were targeted in this study. D Each compound is transformed into atom features and a structure graph for GCN architectures to identify PKFIs. E A visualized explanation was made using the grad-CAM method