Onesum will be at the 60th EUROTOX Congress, “Toxicology without borders,” taking place in Vienna from 13–16 September 2026, with two scientific contributions exploring the integration of Large Language Models into toxicological assessment.
How can Large Language Models support toxicological assessment while maintaining scientific expertise, transparency and expert judgement throughout the process?
This question is central to our work at Onesum. At EUROTOX 2026, Arianna Bassan and Paul Bradley will explore it from two complementary perspectives: the broader integration of LLMs into toxicological risk assessment and their application to comprehensive Target Safety Assessment.
Large Language Models in Toxicological Risk Assessment
Arianna Bassan will present:
Large Language Models in Toxicological Risk Assessment: Strengths, Limitations, and the Expert-Led Workflow
📍 Poster Viewing 1 (PV01) | P03-01 – Artificial Intelligence
The work explores the emerging role of LLMs in Toxicological Risk Assessment (TRA), with a focus on their strengths and limitations and on regulatory expectations for transparency, reproducibility and accountability. An exemplar case study examines LLM-assisted derivation of the Permitted Daily Exposure (PDE) for acetyl tributyl citrate and shows how end-to-end approaches can result in run-to-run variability in point-of-departure selection, modifying factors, cited evidence and PDE values.
The work proposes a modular workflow for the integration of LLMs into TRA, separating data search, information extraction, integrated analysis and summarization, and data interpretation and expert review. This separation is intended to facilitate completeness checks, support reproducibility and aid expert review, helping transform unstructured evidence into expert-driven conclusions and supporting transparency, traceability and expert-led decision making.
LLM-Enabled Workflows for Target Safety Assessment
Paul Bradley will present:
Application of LLM-Enabled Workflows for Comprehensive Target Safety Assessment
📍 Poster Viewing 2 (PV02) | P30-54 – Risk Prediction and Assessment / Risk Assessment Using New Approach Methodologies
This work focuses on Target Safety Assessment (TSA), a structured evaluation of the potential safety risks associated with modulating a biological target. TSA integrates data from a variety of sources, including clinical trials, scientific literature and omics data, to anticipate the potential adverse effects of target modulation. By identifying potential safety liabilities early and proposing mitigation strategies, TSAs help to guide drugs through the discovery and development pipeline.
The work presents an LLM-enabled workflow for TSA evidence generation, integrating third-party data sources via APIs and using retrieval-augmented LLMs with custom prompts to extract, organise and summarise complex evidence. The workflow separates data search, information extraction, integrated analysis and expert review, including validation of citations and interpretations and final sign-off. The resulting output is an expert-validated TSA with a full audit trail.
Human-Centred LLMs for Toxicology
Together, the two contributions explore how LLMs can be integrated into toxicological assessment through structured workflows designed to support transparency, traceability, reproducibility and expert review.
At Onesum, our approach is based on the principle that LLMs should complement and augment human expertise rather than replace it, with structured expert review supporting informed scientific judgement.
Meet the Onesum team at EUROTOX 2026
If you are attending EUROTOX 2026, come and meet Arianna Bassan and Paul Bradley and learn more about their work at their poster presentations.
Interested in learning more?
Get in touch with the Onesum team or visit www.onesumportal.com
Whether you want to explore the Onesum Platform or engage our expert consultants, we’d love to learn about your thoughts and challenges.
contact@onesumportal.com