Researchers develop a machine-learning tool for toxicity analyses that is more consistent in predicting chemical hazards than assays on animals.