What is Onesum?

Toxicological risk assessment is central to ensuring chemical and pharmaceutical safety, but gathering, summarizing, and interpreting the evidence required for these evaluations often demands significant time and effort.

At Onesum, we combine Artificial Intelligence (AI) with the insight of expert toxicologists and a strong foundation in regulatory principles to address these challenges.

This article describes how we integrate AI, especially Large Language Models (LLMs), into toxicological risk assessment while keeping scientific rigor and expert judgment at the center.

Toxicological risk assessment is essential to product development, ensuring the safety of ingredients, biologics, and chemical constituents, including active and inactive components, impurities, excipients, and leachables. In pharmaceuticals and pesticides, these assessments also evaluate risks associated with biological target perturbation. Increasingly, toxicologists must integrate and make decisions on complex and diverse data sources (from internal studies to published literature on both target chemicals and surrogates) placing significant demands on their time and expertise. Producing a single report can take days to months, creating a major bottleneck when scaling assessments across drug target families or classes of chemicals. Gathering exhaustive data, organizing them coherently, and interpreting them accurately all require substantial time and expert attention. A well-structured organization of evidence is, in fact, the foundation for a sound scientific interpretation and the reliability of the final assessment.

The rise of AI, particularly LLMs, offers promise in addressing these challenges by rapidly extracting, synthesizing, and generating human-like outputs from complex documents. However, as widely discussed in the literature, LLMs also present well-recognized limitations. These include, for example, hallucinations, prompt sensitivity, inconsistent outputs, lack of explainability, data privacy concerns, and embedded biases hindering their reliability in regulatory contexts. These issues complicate validation and raise concerns about transparency and trustworthiness in safety-critical applications.

To overcome these challenges, Onesum is advancing scalable, expert-driven solutions that align AI capabilities with the stringent demands of toxicological risk assessment. In partnership with cross-industry collaborators, Onesum is establishing standardized frameworks and workflows for deploying LLMs ensuring consistent, reproducible, and transparent outputs while maintaining toxicologists at the center of decision-making. Its proprietary AI platform operationalizes these workflows, including structured prompt repositories and integrated expert review capabilities, to produce audit-ready reports. Paired with tech-enabled consulting, this model delivers rapid, cost-efficient, and scientifically rigorous assessments at scale.