The Adverse Outcome Pathway (AOP) framework has become central to next-generation risk assessment, yet the development of high-quality, regulatory-ready AOPs remains slow due to the labour-intensive nature of evidence gathering, terminology harmonisation and causal reasoning required for AOP-Wiki submissions. This article outlines how artificial intelligence (AI) - particularly recent advances in natural language processing, ontology alignment and large language models (LLMs) - can accelerate these processes while preserving scientific rigor. The manuscript reviews current bottlenecks in AOP development and evaluates existing AI tools for literature triage, mechanistic evidence extraction, causal-statement structuring, and terminology standardisation. It proposes a practical human-AI hybrid workflow in which AI supports early drafting, evidence synthesis and consistency checks, while expert oversight ensures biological plausibility and alignment with OECD guidance. The article further articulates a community roadmap for AI-enabled AOP development, emphasising low-threshold, high-impact applications that can already be adopted, as well as longer-term needs such as harmonised ontologies, shared training datasets, and integration of AI-assisted functions into AOP-Wiki 3.0. Initiatives such as AI4AOP are highlighted as real-world pilots demonstrating the feasibility of accessible, reproducible AI workflows for non-expert users. Together, these efforts aim to scale up high-quality AOP creation, enhance transparency, and strengthen the evidence base underpinning regulatory decision-making.

SONG You;  KUMAR Vikas;  TANABE Shihori;  VILLENEUVE Dan;  WITTWEHR Clemens; 

2026-07-24

AMER CHEMICAL SOC

JRC145118

1520-5851 (online),   

https://doi.org/10.1021/acs.est.6c05148,    https://publications.jrc.ec.europa.eu/repository/handle/JRC145118,   

10.1021/acs.est.6c05148 (online),