The use of Bayesian methodology in the development and validation of a tiered assessment approach towards prediction of rat acute oral toxicity

Journal article

EPAA’s ‘NAM Designathon 2023’ challenge for human toxicity sought to identify a classification system capable of categorising chemicals based on their bioactivity and bioavailability properties determined using non-animal methodologies (Worth et al. 2025). The proposal is made to classify chemicals into three levels of concern: low concern could be used without restriction, medium concern requiring assessment to establish safe use levels and high concern being candidates requiring risk management (Berggren and Worth in Regul Toxicol Pharmacol 142:105431, https://doi.org/10.1016/j.yrtph.2023.105431, 2023). We developed a NAMs based classification system for “human systemic toxicity” mainly focussed on repeat dose toxicity, similar to the assessment carried out in classification for ‘Specific Target Organ Toxicity—Repeated Exposure’ (STOT-RE) based on ECETOC’s Tiered Approach integrating three lines of evidence: In silico predictions, In vitro bioavailability and PBK modelling, In vitro bioactivity assays. The first stage employed an in silico approach, covering several toxicity endpoints across various (Q)SAR in silico models to identify indicators of toxicity. Bioavailability was categorised by simulating 14-day plasma Cmax predictions for a standard dose level using three TK models (Firman et al. in Arch Toxicol 96:817–830, https://doi.org/10.1007/s00204-021-03205-x, 2022). Bioactivity was categorised using a matrix with potency and severity. In vitro data were obtained from ToxCast. Potency makes use of dose response AC50 values. Severity categorisation is based on consideration of the adverse effects associated with the assays. 12 chemicals have been assessed through the framework. Overall, we have demonstrated that the matrix suggested by the EPAA Designathon can be used to categorise chemicals into three different levels of concern but there are areas still to be explored especially for the range of assays used, the framework categorisation being defined, and how such a matrix would fit into a tiered approach, pragmatically, including targeted in vivo studies

Authors: Firman, JW; Cronin, MTD; Rowe, PH; Semenova, E; Doe, JE

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Springer
• Journal: Archives of Toxicology
• Volume: 96
• Issue: 3
• Pages: 817-830
• Publication date: 2022-01-16
• DOI: 10.1007/s00204-021-03205-x
• EISSN: 1432-0738
• ISSN: 0340-5761
• Language: English
• Keywords: Medicine; Scope (computer science); Hazard; Risk analysis (engineering); Computer science; In silico; Machine learning; Artificial intelligence; Biology; Bayesian inference; Set (abstract data type); Robustness (evolution); Bayesian probability; Computer security; Inference; Risk assessment; Data mining