Integrating in silico ADME modeling with Tox21 toxicity profiling to support compound prioritization

Abstract Toxicity remains a leading cause of late-stage attrition in drug development and is heavily influenced by a compound’s ADME (Absorption, Distribution, Metabolism, and Excretion) properties. The ability of a molecule to reach its target tissue, persist in circulation, or undergo bioactivation fundamentally shapes its toxicological profile, highlighting the need to integrate pharmacokinetics into toxicity assessment. The Tox21 program, a multi-agency collaborative initiative among NCATS, EPA, FDA, and DTT/NIEHS, has advanced mechanism-based high-throughput screening to assess over 8,000 chemicals across more than 90 toxicity-relevant assays. While these data provide critical mechanistic insights, real-world risk assessment requires contextualizing toxicity with ADME characteristics such as solubility, metabolic stability, and permeability. To address this need, NCATS developed the ADME@NCATS platform, which provides experimentally derived ADME datasets and machine learning models trained on tens of thousands of compounds. In this study, we evaluated the applicability of ADME@NCATS models to selected compounds from the Tox21 10K library using predictions of rat liver microsomal stability, aqueous solubility, PAMPA-gastrointestinal permeability, and PAMPA-blood–brain barrier permeability, followed by experimental evaluation on selected compounds. Chemical space and structural-similarity analyses were used to assess the relationship between the Tox21 compounds and the ADME@NCATS training datasets. The resulting ADME predictions provide complementary information on compound disposition that can be integrated with pharmacological, toxicological, and other biological data to inform compound selection and prioritization throughout the drug discovery process.