Fig. 4From: Integrating concept of pharmacophore with graph neural networks for chemical property prediction and interpretationEffects on predictive performance that RG pooling brings to basis models. a AUC gains based on random splitting datasets. b AUC gains based on scaffold splitting datasets. Most of the AUC gains in both subfigures are positive, which means that the RG pooling is helpful to improve predictive performance of modelsBack to article page