抄録
High drug failure rates within drug development remains a significant concern. To predict human safety liabilities integration of new approach methodologies (NAMs) into early drug discovery has gained significant momentum. The use of NAMs allows for greater understanding of drug mechanisms and prediction of toxicity potential earlier in the discovery process. By leveraging human-relevant high-throughput in vitro assays, such as 3D organ systems, and transcriptomic profiling, chemical series can be assessed and triaged based on their toxicity profile. In this study, we outline an early de-risking strategy for drug-induced liver injury (DILI) using six NAMs covering the key hallmarks of DILI. A reference compound set of 130 DILI negative and positive compounds (true positive = 70, true negative = 60) was utilised. The NAMs included the Glu/Gal, BSEP, HepG2 imaging, Mitochondrial Stress Test, HepaRG liver spheroid imaging and whole genome transcriptomics. The data from the first five NAMs was integrated with a XGBoost model to predict DILI and the transcriptomics approach utilised AI modelling. The XGBoost model as an early DILI safety strategy gave high predictive value for DILI concordance (Sensitivity = 81%, Specificity = 83% and Accuracy = 82%). However, utilising transcriptomics improved the risk assessment and favourable compound selection (Sensitivity = 81%, Specificity = 95% and Accuracy = 88%). A similar observation was observed using NAMs to de-risk cardiotoxicity. This study illustrates the benefits of using NAMs to de-risk compounds in early drug discovery.
