DeepCyte launched DeeTox Atlas, a single-cell metabolomic reference atlas designed to map drug toxicity mechanisms. The company said the foundation dataset is built from two independent single-cell metabolomics perturbation studies, spanning roughly 100 toxicant compounds, 300,000 cells, and about 500 metabolites per cell. DeepCyte claims it trains a foundation model to predict toxicity mechanisms for compounds it has never measured and ties each compound to a curated hierarchy of toxicity mechanisms anchored to established Adverse Outcome Pathways (AOPs). The goal is to shift toxicity testing from reactive experiments to mechanism-based, predictive AI earlier in drug discovery. Enterprise pilot programs with global pharmaceutical companies are expected to start in the coming months, positioning the atlas as a scaling dataset to reduce wet-lab dependence while improving predictive performance. For translational and safety teams, the product targets a known bottleneck—linking compound chemistry to mechanism-resolved safety liabilities—using single-cell resolution to capture rare cell subpopulations.