OS03-09 Surveying the Landscape: A Modular Generative Artificial Intelligence Workflow to Identify NAMs for Systemic Toxicity
Year: 2025
Type: article
Abstract: To identify the areas of systemic toxicity with the greatest need, and which present the best opportunities for human-relevant model (i.e., new approach methodologies [NAMs]) development, standardization, and implementation, we conducted a landscape analysis to collect information on ongoing efforts in the NAMs space. This type of analysis traditionally requires the collection, manual review, summary, and synthesis of an extensive literature base that requires hundreds of hours to complete. To increase both speed and efficiency, we utilized a reproducible workflow that incorporates multiple computational tools including generative artificial intelligence (GenAI) to quickly summarize a large literature database. Our integrated approach coupled subject matter expertise with sorting and extraction algorithms to provide a comprehensive overview of the state of the science for NAMs used for, or potentially useful for, the assessment of systemic toxicity of cosmetics. To identify potentially relevant studies, we conducted a literature search with keywords related to NAMs across three major topic areas: in silico , in chemico, and in vitro . We prioritized studies by coupling supervised clustering, topic extraction and keyword analysis algorithms. These methods led to the prioritization of 8,418 studies. To identify relevant studies, subject matter experts were employed in conjunction with active machine learning to identify relevant studies that were then summarized via GenAI. Given the objective was to provide sufficient coverage of the landscape to both address pragmatic, near-term needs, as well as shaping the future of how safety assessments are performed, we designed prompts to characterize the current and past systematic efforts directed towards developing and refining NAMs, including both success stories, scientific and technical challenges, and roadblocks to wider adoption. Our analysis identified 3,010 peer-reviewed publications and 38 consortium websites cataloguing NAMs that were applicable to the cosmetics regulatory process, from hazard-focused endpoints to exposure-based waiving of studies altogether. Additionally, they covered the full spectrum of maturity, from those approaches that show promise at the research and development phase to fully validated approaches ready for immediate regulatory use. To the extent available, we identified the molecular/cellular endpoints associated with each NAM and the reference dataset that was used to develop and/or evaluate usefulness and limitations across a total of 60,960 endpoints. The landscape also captured opportunities to validate mature NAMs to support their regulatory use and market adoption. These results will support the development and identification of NAMs to be included in frameworks for assessing systemic toxicity potential.
Source: Toxicology Letters
Authors: D. Allen, E. Martin, J. Hamm, J. Wignall, K. To, T. Feiler, C. Lemeris, P. Kukic
Institution: Unilever (United Kingdom)
Language: English
Subfield: Small Animals
Field: Veterinary
Domain: Health Sciences
Open Access status: hybrid
Estimated APC paid: $4,270
OpenAlex ID: W4414075482