Fig. 6: ToxPredictor outperforms state-of-the-art prediction models in accuracy and scalability.
From: A large-scale human toxicogenomics resource for drug-induced liver injury prediction
A Balanced accuracy versus scale (number of DILI + /− compounds) across published preclinical models. ToxPredictor achieves the highest performance among both in vitro and in silico methods, outperforming high-content imaging, cytotoxicity, bioactivation-based and structure-based models, demonstrating robust performance even when scaled to hundreds of compounds. B Head-to-head comparison on overlapping compounds shows consistent performance gains over other studies. Models with >80% specificity (marked with *) support credible preclinical de-risking. Results highlight transcriptomics-based toxicogenomics as a leading strategy for mechanistically informed, scalable DILI prediction.
