The concordance of biomarker genes between Tox-generative adversarial network (GAN) model versus the model from the real data: Percentage of overlapping genes between the top-ranking genes derived from liver necrosis deep neural network (DNN) models based on generated transcriptomic profiles and real ones were compared where the generated transcriptomic profiles were from Tox-GAN_intensity. The important genes were extracted from DNN models using the Deep Approximate Shapley Propagation method.