Key points are not available for this paper at this time.
Drug-Induced Liver Injury (DILI) remains one of the most critical challenges in drug development, causing patient safety concerns, clinical trial failures and drug withdrawals. We introduce ToxPredictor, a toxicogenomics framework combining RNA-seq data from primary human hepatocytes with pharmacokinetic data to predict dose-resolved DILI risks and safety margins. At its core is DILImap, an RNA-seq library tailored for DILI research, comprising 300 compounds at multiple concentrations. ToxPredictor achieves 88% sensitivity at 100% specificity in blind validation, outperforming state-of-the-art methods. It flagged recent phase III clinical failures, including Evobrutinib, TAK-875, and BMS-986142, overlooked by animal studies. Beyond prediction, ToxPredictor provides mechanistic insights into hepatotoxic pathways, enabling early de-risking and actionable safety decisions. Unlike single-endpoint readouts—even from 3D models—transcriptomics offers a multi-dimensional system-level view of hepatocyte responses, capable of detecting diverse DILI mechanisms not captured by conventional assays. Scalable, actionable, and integrated into a broader AI/ML drug discovery platform, this work establishes toxicogenomics as a promising tool for developing safer therapeutics and addressing one of the most pressing challenges in toxicology. Drug-induced liver injury is a major cause of patient harm, trial failures, and drug withdrawals. Here, the authors show that a toxicogenomics resource of 300 drugs enables the prediction of liver injury with 88% sensitivity at 100% specificity and reveals mechanisms for safer drug development.
Bergen et al. (Thu,) studied this question.