Cellarity published an AI toxicogenomics model, ToxPredictor, trained on DILImap — a transcriptomics library from primary human hepatocytes profiling 300 compounds across doses — to predict dose‑related drug‑induced liver injury (DILI). Reported validation showed 88% sensitivity at 100% specificity and retrospectively flagged Phase III safety failures missed in animal studies. The model and validation data were shared open‑source for non‑commercial use in Nature Communications. DILI prediction is a late‑stage safety bottleneck that can sink drug programs; mechanistic transcriptomics combined with machine learning could change preclinical safety triage and dose selection.