Cellarity published a toxicogenomics‑based AI model, ToxPredictor, that leverages a DILImap library of transcriptional signatures from 300 compounds to predict drug‑induced liver injury (DILI). Validation reported 88% sensitivity at 100% specificity in identifying DILI‑linked compounds and the tool retrospectively flagged Phase III clinical safety failures missed by animal studies. The model uses an active learning framework and the company opened the dataset for non‑commercial use. Cellarity positions the tool to help drug discovery teams identify dose‑related liver risks earlier and guide go/no‑go decisions, potentially saving late‑stage failures. ToxPredictor is an example of AI applied to safety prediction where transcriptomics contextualizes mechanism; sponsors and CROs may adopt such models to de‑risk portfolios prior to expensive clinical testing.