AI Tools in ONTOX
This section highlights AI tools used or developed within the ONTOX project.
AI Tools in ONTOX
Integration of AI tools into new approach methods (NAMs) for toxicology represents a paradigm shift in chemical safety assessment.
Sysrev
Sysrev is a human-in-the-loop evidence review platform that accelerates systematic literature review and data extraction workflows. The platform enables teams to collaboratively screen, annotate, and extract structured data from scientific literature with integrated quality control and validation features. Sysrev has supported over 20,000 users across academic, government, and industry applications. Within ONTOX, Sysrev facilitated collaborative evidence synthesis, automated data extraction using large language models, and standardized data curation workflows. The platform supported 62 ONTOX projects including publications on genetic susceptibility in Parkinson’s disease risk assessment and drug-induced fatty liver disease ontogeny.
Website: https://sysrev.com
ToxIndex
ToxIndex is an agentic platform for regulatory toxicology that integrates data sources, predictive models, and analytical workflows to automate risk assessment processes. The system orchestrates multiple computational tools to address chemical safety evaluation challenges, supporting applications from hazard identification through regulatory dossier generation. ToxIndex combines curated toxicology data infrastructure with AI-powered workflow automation to reduce assessment timelines while maintaining scientific rigor and audit trails suitable for regulatory review. Within ONTOX, ToxIndex capabilities supported data gap analysis, automated identification of relevant testing strategies, and demonstration of the OPRA (ONTOX Probabilistic Risk Assessment) framework at the ECETOC-VHP4Safety-ONTOX workshop in Brussels (October 2025).
Website: https://toxindex.com
ToxTransformer
ToxTransformer is a proprietary multi-property prediction model for chemical toxicity assessment. The model predicts multiple toxicological properties simultaneously from molecular structure, enabling conditional predictions where known properties improve accuracy for unknown endpoints. Trained on the ChemHarmony dataset (117 million chemicals, 254 million chemical activity records), ToxTransformer predicts over 4,000 toxicological, ADMET, and environmental properties. The model’s architecture enables cross-property transfer learning and counterfactual analysis to support testing prioritization and read-across approaches. Within ONTOX, ToxTransformer was integrated into the ToxIndex platform to support data gap filling, hazard prediction, and identification of relevant OECD in vitro assays for missing endpoints.
acknowledgment
These tools were developed by Insilica with US federal grant funding (NSF and NIH SBIR) and made available to ONTOX consortium partners to support the project’s research objectives in next-generation risk assessment.