Objective: To develop virtual animal models to simulate animal study results using artificial intelligence (AI).
Introduction: Testing data from animal models provide crucial evidence for the safety evaluation of chemicals. These data have been an essential component in regulating drug, food, and chemical safety by regulatory agencies around the world. As a result, a wealth of animal data are available from the public domain and other sources. As the toxicology community and regulatory agencies move towards a reduction, refinement, and replacement (3Rs principle) of animal studies, we are exploring an AI-based generative adversarial network (GAN) architecture to learn from existing animal studies to generate animal data without conducting additional animal experiments.
Approaches: We are developing a GAN method that can generate animal-study data which otherwise would have to be obtained with actual animal studies. In a pilot study, we demonstrated that synthetic data from AnimalGAN for toxicogenomics, hematology, and clinical chemistry has the potential to be used in toxicity assessments, mechanistic studies, and biomarker development similar to actual experimental data is used.
Potential Impact: Conventional animal studies can be expensive, time-consuming, labor-intensive, and raise ethical concerns. AnimalGAN is an AI-based suite to generate specific animal-study datasets for new and untested chemicals by learning from legacy animal-study data. Thus, it could supplement animal studies by screening new and untested chemicals to guide and refine subsequent animal experiments in risk assessment and safety evaluation.
Tox-GAN: An Artificial Intelligence Approach Alternative to Animal Studies—A Case Study With Toxicogenomics.
Chen X., Roberts R., Tong W. et al.
Toxicological Sciences. 2022,186:242-259.