ToxReason is a benchmark for evaluating whether large language models can reason about chemical toxicity through valid biological mechanisms not just predict toxic outcomes. Grounded in the Adverse Outcome Pathway (AOP) framework, it assesses organ-level toxicity reasoning across multiple organs by requiring models to trace the mechanistic chain from Molecular Initiating Event (MIE) to Adverse Outcome (AO).
We're actively cleaning up the code and documentation. Full release coming soon.
Please cite related papers/blogs using this BibTeX if you find this useful for your research and applications.
@inproceedings{park2026toxreason,
title={ToxReason: A Benchmark for Mechanistic Chemical Toxicity Reasoning via Adverse Outcome Pathway},
author={Park, Jueon and Jang, Wonjune and Kim, Chanhwi and Park, Yein and Kang, Jaewoo},
booktitle={Findings of the Association for Computational Linguistics: ACL 2026},
pages={19542--19564},
year={2026}
}