Fig. 3: AnimalGAN for toxicity assessment. | Nature Communications

Fig. 3: AnimalGAN for toxicity assessment.

From: A generative adversarial network model alternative to animal studies for clinical pathology assessment

Fig. 3

a A framework for comparing toxicity assessment outcomes between clinical pathology measurements generated by AnimalGAN and those from real rat experiments conducted under identical treatment conditions. Clinical pathology measurements were generated by AnimalGAN for each treatment condition (i.e., compound/time/dose). Then, each generated clinical pathology measurement and its corresponding real one (i.e., treated group) were analyzed against their matched controls to establish a statistically significant toxicity outcome. If both real and synthetic data lead to conclude the same toxicity outcomes, we consider that an agreement or “consistency” is established between experiment and AnimalGAN. The consistency for b hepatotoxicity and c nephrotoxicity-related clinical pathology measurements between generated data and their corresponding animal testing data in the test set. The consistency results for all the 38 clinical pathology measurements can be found in the Supplementary Table 1.

Back to article page