Cellarity, a clinical-stage biotechnology company developing Cell State-Correcting therapies through AI and multi-omics integration, has published a groundbreaking study in Nature Communications detailing a new framework for predicting and analyzing drug-induced liver injury (DILI). The research introduces ToxPredictor, an AI-driven model that leverages toxicogenomics to assess the risk of DILI during drug development. This is a major step forward in addressing one of the most persistent challenges in pharmaceuticals: hepatic safety issues that often go undetected in preclinical testing and lead to clinical trial failures or even market withdrawals. Animal models fail to predict DILI in up to half of investigational drugs, highlighting the urgent need for more accurate human-relevant safety assessments. To address this, Cellarity developed DILImap, a comprehensive transcriptomics library derived from primary human hepatocytes. DILImap captures the gene expression profiles of 300 compounds linked to DILI across multiple doses, making it the largest publicly available toxicogenomics dataset for liver injury modeling. This dataset forms the foundation of ToxPredictor, which uses machine learning to analyze complex molecular patterns and predict DILI risk with high precision. In blind validation tests, the model achieved 88% sensitivity at 100% specificity—outperforming more than 20 standard preclinical safety models. Notably, it successfully identified safety failures from phase 3 clinical trials that had previously evaded detection in animal studies. The framework excels at uncovering diverse DILI mechanisms, including mitochondrial dysfunction, oxidative stress, immune activation, and metabolic disruptions—many of which are missed by conventional assays or even advanced 3D cell models. Unlike single-endpoint tests, transcriptomics provides a high-resolution view of the entire molecular landscape, enabling deeper insights into how drugs affect liver cells. This allows researchers to not only predict toxicity but also understand the underlying biological pathways, supporting better decisions on compound safety margins and enabling earlier de-risking of drug candidates. In a significant move toward open science, Cellarity has released ToxPredictor, DILImap, and all validation data as open-source resources at www.dilimap.org. This initiative aims to accelerate innovation in drug safety by enabling collaboration across academia, industry, and regulatory agencies, potentially reducing reliance on animal testing and transforming predictive toxicology. Parul Doshi, Cellarity’s Chief Data Officer, emphasized that applying AI to toxicogenomics offers a powerful path toward more efficient, cost-effective, and safer drug development. “Our model provides deep insights into liver toxicity mechanisms,” she said, “and holds great promise for improving patient safety.” Founded in 2019 by Flagship Pioneering, Cellarity is advancing a novel drug discovery platform that uses dynamic AI and transcriptomics to correct dysfunctional cell states. Its lead candidate, CLY-124, is in Phase 1 trials for sickle cell disease, targeting a novel globin-switching mechanism. Additional programs are in development for hematology, immunology, and metabolic diseases, including a collaboration with Novo Nordisk focused on metabolic dysfunction-associated steatohepatitis (MASH). The publication of this research marks a pivotal moment in the integration of AI and multi-omics in drug safety, offering a scalable, human-relevant solution to a long-standing industry challenge. By making its tools publicly available, Cellarity is helping to set a new standard for transparency and innovation in pharmaceutical development.