Continuing Education

Course Book

Chair(s):
Kan Shao, Indiana University; and Chao Ji, ATSDR.

Benchmark dose (BMD) methodology is a quantitative dose-response modeling approach used to identify points of departure associated with defined levels of response. It has been widely accepted as a preferred alternative to the traditional no-observed-adverse-effect level (NOAEL) approach because it uses more information from the dose-response data, is less dependent on the specific dose spacing used in a study, and provides a stronger basis for characterizing statistical uncertainty.

As chemical risk assessment increasingly incorporates diverse evidence streams, including animal toxicology studies, epidemiological data, high-throughput screening results, genomic data, and new approach methodologies (NAMs), there is a growing need for harmonized BMD modeling strategies that can be applied consistently across different data and study types in terms of modeling procedures and BMD estimation. This course will provide a practical introduction to BMD modeling and its application in risk assessment decision-making. The course will include two presentations: a presentation on Bayesian Benchmark Dose Modeling (BBMD) approaches across data and study types and a presentation on the use of BMD modeling in public health reference value derivation. These presentations will be followed by a question-and-answer period.

The first presentation, “Bayesian Benchmark Dose (BMD) Modeling Across Data and Study Types: A Harmonized Strategy for Chemical Risk Assessment,” will introduce the BMD methodology and demonstrate how the BBMD system can be used to analyze dichotomous, continuous, and categorical dose-response data from toxicological, epidemiological, and genomic studies using harmonized procedures for model fitting, goodness-of-fit evaluation, model selection, uncertainty characterization, and model-averaged BMD analysis in a common Bayesian framework. The second presentation, “From Data to Decisions: An Introduction to Benchmark Dose Modeling in Public Health,” will use the derivation of human health reference values, including ATSDR minimal risk levels, as a case study to show how BMD-derived points of departure are integrated with duration and dosimetry adjustments, uncertainty factors, animal and human data, NAMs, and emerging quantitative uncertainty methods to support public health decision-making.

Chair(s):
Beau Rollins, Intrinsic Medical Group; and Vivek Lawana, NAMSA.

Neurological medical devices are generally classified as high-risk medical products due to their direct and prolonged interaction with the central and/or peripheral nervous system, long-term implantation, and potential for severe irreversible harm. Per the latest ISO 10993-1 standard, medical devices or their constituents that have direct or indirect contact with tissues of the central or peripheral nervous system or cerebrospinal fluid (CSF) require explicit evaluation for potential neurotoxicity, including both local neurotoxicity at the tissue–device interface and systemic neurological effects. As a result, neural-based devices—and other devices with credible neural exposure—often require more extensive and carefully integrated preclinical safety evaluation strategies. This course will bring together experts in regulatory consulting, biocompatibility, toxicology and risk assessment, and regulatory review to provide a practical, experience-based discussion on how to implement and interpret local, systemic, and neurotoxicity-relevant endpoints for neural-based medical devices. Particular emphasis will be placed on large-animal implantation studies, where assessment of local tissue response, systemic exposure, and the neural microenvironment must be considered together to support meaningful safety conclusions.

The course will begin with a presentation from a regulatory consultant, who will outline global regulatory expectations for high-risk neurological devices and discuss how regulators interpret ISO 10993-1 requirements related to neurotoxicity. This talk will highlight common regulatory questions, expectations for preclinical safety evidence, and the importance of early alignment between device design, intended use, and biological safety strategy. The second presentation will focus on the practical implementation of combined local and systemic endpoints in large-animal studies, drawing on ISO 10993-6 and ISO 10993-11. Study design considerations—including animal model selection, implantation duration, endpoint selection, and integration of systemic assessments with local neural tissue evaluations—will be discussed, with attention to avoiding redundant or non-informative testing. The third speaker will address toxicological risk assessment for neuronal medical devices, emphasizing how large-animal implantation data are incorporated into an overall biological risk assessment under ISO 10993-1. This presentation will discuss the interpretation of findings within the context of the neural microenvironment, including inflammation, foreign body response, tissue remodeling, and long-term adaptation, as well as approaches to managing uncertainty and supporting benefit–risk conclusions. The course will conclude with perspectives from a regulatory expert representing a Japanese notified body, who will share insights into regional expectations for evaluating neurotoxicity, implantation findings, and systemic data. This presentation will address common areas of regulatory focus and opportunities for harmonizing global approaches to safety evaluation of high-risk devices with neural exposure considerations.

Collectively, this course will provide attendees with a practical, end-to-end understanding of how ISO 10993-1 neurotoxicity considerations can be effectively addressed through integrated study design, large-animal implantation data, and risk-based evaluation, offering real-world guidance for navigating the heightened safety expectations associated with neural-based medical devices.

Chair(s):
Shannon Bell, RTI International; and Paul Whaley, Lancaster Environment Centre, Lancaster University, United Kingdom

For research to make a difference, it must be discoverable and interpretable by the people who need to use it. The explosion of research publications and the ways research is accessed have changed dramatically over the past few years. From “science by press release,” social media commentaries, and short-form video to preprint archives and presentation repositories, the sharing of research can take many forms to target different audiences.

Despite advances in data sharing, publication access, and analytic tools, substantial gaps remain between producing research and ensuring data are taken up and applied. Many studies are not readily discoverable, lack sufficient reporting for interpretation, or fail to provide data in formats that enable reuse. Competing for space in an attention economy feels stressful yet essential. This gap between research products and their use limits the impact of scientific investments and slows translation into regulatory, clinical, and public health decision-making.

Designed for early- and mid-career investigators, this interactive course equips participants with practical strategies to go beyond publishing solid data by also identifying their audiences and what they need, designing feasible strategies for reaching their audiences, and creating research outputs that not only present the science, but are easier to find, understand, and use. Participants will work through a “get noticed” playbook for research dissemination with course facilitators, learning how relatively small changes in research summaries, reporting practices, terminology, and data sharing can substantially increase the visibility, credibility, and real-world impact of their work. We will work through the playbook using audience personas (general public, decision-makers, secondary data users, and peer researchers) to help participants effectively target their outreach.

The course will help researchers think more strategically about how to reach and serve key audiences, including other scientists, regulators, and the public. Participants will leave with a practical framework and immediately applicable strategies for increasing the discoverability, interpretability, and usability of their research, skills that can enhance both the impact of their science and their professional visibility in a changing landscape.

Chair(s):
Shagun Krishna, Physicians Committee for Responsible Medicine (PCRM); and Alexandre Borrel, Sciome, LLC.

Computational prediction tools have transformed how toxicologists assess chemical hazards, enabling faster, more cost-effective, and human-relevant safety decisions without reliance on animal testing. Chief among these is the OECD QSAR Toolbox, a free, globally used regulatory decision-support platform developed by the Organisation for Economic Co-operation and Development (OECD), which supports chemical profiling, chemical category formation, and read-across predictions. Despite its growing prominence in regulatory submissions worldwide, many toxicologists have limited hands-on experience with the tool and uncertainty about how to use its outputs confidently in real regulatory contexts.

This Continuing Education course is designed to close that gap. It offers a structured, progressive learning experience that moves participants from foundational principles of structure-activity relationships and into a live, guided walkthrough of the OECD QSAR Toolbox, with explicit connections to regulatory frameworks governing how these predictions are evaluated and accepted by agencies in the United States, Europe, and internationally.

The course unfolds in two blocks. The first establishes the conceptual and regulatory grounding participants need: how structure-activity models are built and validated, how standardized modeling pipelines connect to regulatory documentation requirements, and how international frameworks guide regulators in evaluating the scientific validity, transparency, and reliability of computational predictions. This block is delivered by experts spanning computational science, cheminformatics, and regulatory toxicology, ensuring that participants understand not just how the tools work but also how regulators apply these frameworks.

The second block puts that knowledge to work. Participants will engage in live, hands-on sessions co-facilitated by scientists with direct experience developing and implementing the OECD QSAR Toolbox in regulatory contexts. Working through real chemical examples, participants will perform chemical profiling, build chemical categories, select analogs, and execute read-across predictions, the same workflows used in actual regulatory submissions. The emphasis throughout is on making decisions transparently and defensibly, not on treating computational outputs as automatic answers.

This course is designed for toxicologists across all sectors, including regulatory agencies, industry, consulting, and academia, at all career stages. Attendees will leave with both the conceptual grounding and the practical skills needed to incorporate structure-activity tools into their safety assessment work with confidence.

Chair(s):
Justin Teeguarden, Pacific Northwest National Laboratory; and Katrina Waters, Pacific Northwest National Laboratory.

Like most scientific fields, toxicology is becoming increasingly interdisciplinary, requiring larger, more complex, and more distributed teams of domain experts to compete effectively and deliver high-impact outcomes. The success of these teams depends fundamentally on the caliber of leaders guiding their teams. Outside of science, the impact of leadership development on career trajectory is consistently documented by major research firms like Gallup, LinkedIn, and Harvard Business Review. Modern data highlights a “Leadership ROI” that benefits both an individual’s promotion potential and team and organizational productivity. Yet, formal training in the core competencies of leading the complex scientific teams common in toxicology is largely absent.

This highly participatory session addresses this deficit by introducing participants to the “Science of Team Science,” an evolving, empirical field that treats leadership as a technical competency that can be taught, learned, honed, measured, and optimized for greater scientific productivity and career development. From this introduction, the session moves to recent research on core competencies for team science, highlighting the centrality of highly effective communication to each competency. To move these theories into the field, the workshop introduces a practical, data-driven instrument known as the Social Styles Model (Merrill & Reid, Wilson Learning). Framed as “Observational Phenotyping,” this model and the supporting field guide allow science leaders to quickly discover communication preferences with the same rigor they use in their research. By understanding these styles, leaders can tailor communication with individuals and whole teams, adapting their engagement strategies to reduce relationship tension and enhance interdisciplinary ideation, creativity, team cohesion, and productivity. Participants will learn to identify the four primary social styles and their corresponding “Conflict Sequences,” gaining the ability to anticipate, prevent, or resolve the stylistic clashes that often derail interdisciplinary research teams. Facilitated by Senior Fellows in toxicology and an Executive Coach, the session concludes with a cohort-based exploration of real-world cases of the participants’ choosing. Working with workshop stewards, attendees will apply these observational tools to diagnose team friction and engineer actionable solutions for their own interdisciplinary teams. The overarching goal of the course is to provide attendees with a practical understanding of specific leadership competencies from the emerging field of the Science of Team Science and a new element for their personal “Leadership Toolkit” that allows them to lead their teams more effectively by identifying and adjusting to the unique communication styles of those on their teams.

Chair(s):
Stephen Edwards, UL Research Institutes’ Chemical Insights; and Annie Jarabek, US EPA.

The increased volume and diversity of data used to support environmental risk assessment has intensified reliance on computational approaches to assess toxicity and exposure. The diversity of data also drives the need for reliable interoperability as multiple models are typically needed to describe the full source-to-outcome continuum. This requires an environmental health language that is understood by humans and computers alike.

The Environmental Health Language Collaborative (EHLC) was formed in 2021 to address this need. Early results from this effort were presented in a Workshop Session during the 2024 SOT Annual Meeting. Now, the lessons learned from this collective effort are being provided as a Continuing Education (CE) course, which shows environmental health scientists how data standards and “standardized language” (controlled terminologies and ontologies) make evidence easier to integrate, reuse, and analyze across toxicology, exposure science, and environmental epidemiology to support risk assessment. This course introduces the practical value of data standards, standardized language, and emerging computational tools—including knowledge graphs and machine learning approaches, such as large language models (LLMs)—for evidence integration and cross-study analysis.

The course will be framed around a case study using the integrated science assessment of fine particulate matter (PM2.5) and respiratory outcomes. The first presentation will provide an overview of the risk assessment process and demonstrate the need for a common, computable language to enable coherent data integration, harmonization, and modeling across exposure, epidemiology, animal toxicology, mechanistic studies, and new approach methodologies (NAMs) when characterizing chemical risk. It will also introduce the PM2.5 case study and put the material presented by subsequent speakers in this context. The second presentation will show how ontologies and common data elements can promote data integration and serve as the building blocks for ensuring that toxicology and environmental epidemiology data are FAIR (Findable, Accessible, Interoperable, Reusable) using PM2.5 and respiratory outcomes as an example. The third presentation will continue our consideration of the case study through the integration of toxicology and exposure data in exposomics studies with an emphasis on airborne pollutants such as PM2.5. The presentation will show how semantic data models and knowledge graphs provide data in a form suitable for computational modeling. The fourth presentation focuses on data harmonization within environmental epidemiology, where differences in variable definitions, units, instruments, timing, and metadata often block cross-study comparisons. The impact of data harmonization, driven by data and terminology standards, on our PM2.5 case study will be revealed using examples from the Human Health Exposure Analysis Resource (HHEAR) repository. The final presentation will incorporate the components from the previous talks and display how the concerns raised at the start of the course can be addressed through a common, computable language that incorporates toxicology, exposure science, and epidemiology into a quantitative mechanistic model linking PM2.5 to decreased lung function. This example will also highlight how emerging machine learning tools (including large language models) may assist with data alignment and reuse. Each presentation will include a lecturer–attendee interactive session where the audience will be asked to participate in solving the problem.

Chair(s):
Barbara Wetmore, US EPA; and Katie Paul Friedman, UL Research Institutes’ Chemical Insights.

New approach methods (NAMs), including in silico and in vitro screening methods, can decrease the time required for chemical risk assessment. Applied NAM research activities have matured from mere demonstrations to strategic activities to garner acceptance for specific decision contexts regarding chemical safety. This includes development of tiered, fit-for-purpose frameworks, operationalizing open-source NAM databases and tools via user-friendly interfaces, and cross-sector case studies presenting implementation strategies that foster understanding and confidence in these approaches. These efforts have resulted in the infusion of NAMs into varying decision-making strategies that are actively pursued internationally for specific toxicology applications such as prioritization, weight of evidence, and integrated approaches to testing and assessment.

This course will cover recent advances, such as extending the application of threshold of toxicological concern for volatiles, and presenting workflows that incorporate metabolism within NAMs. Each presentation has been structured to include either hands-on exercises or practical demonstrations. The ultimate goal of this course is to provide the attendees with not only technical guidance for workflow execution, but access to data and code resources for pragmatic application.

The course will be introduced with a brief overview of NAM-based testing frameworks and how these frameworks follow a similar workflow: use relevant in silico predictions, gather all data, and focus on data gaps for additional NAM-based testing. Each speaker in this course will speak to one or more of these unifying elements across NAM-based testing frameworks and leave attendees with actionable data or strategies for NAM-based chemical assessment. The confirmed speakers include subject-matter experts with many years of experience in the development and application of these tools and with this course aim to educate researchers on (1) advances in application of the threshold of toxicological concern (TTC) for challenging airborne compounds; (2) available databases of in vivo hazard and in vitro points of departure and application programming interfaces that enable risk-based prioritization and in vitro–in vivo concordance evaluations; (3) recent advances in physiologically based kinetics, in vitro distribution modeling, and in vitro–in vivo extrapolation; and (4) data analysis tools that incorporate exposure with bioactivity for rapid risk prioritization. The course will then transition to two case studies: one describing recent advances in incorporating metabolism with NAM high-throughput profiling data and guidance for data integration, interpretation, and implementation, followed by an industry-based example of operationalizing NAM data and tools for decision-making. Time will be allotted for direct interactions between the attendees and speakers as they walk through examples, facilitating a more personal interaction to address the attendees’ specific research questions and needs.

Chair(s):
Hollie Skaggs, Amgen; and Heather Estrella, Amgen.

Artificial intelligence and machine learning (AI/ML) are increasingly being incorporated into pharmaceutical research and development; however, their practical application in toxicology remains variable. While many toxicologists are familiar with AI/ML concepts, there is often a gap between awareness and implementation. This course is designed to address this gap by providing a structured, applied introduction to how AI/ML approaches can support real-world toxicological decision-making.

The course focuses on a common and challenging problem in pharmaceutical toxicology: the interpretation and translation of gastrointestinal (GI) toxicity findings observed in nonclinical dog studies to potential human risk. GI findings—including vomiting, diarrhea, decreased food intake, and histopathological evidence of mucosal injury—are frequently observed in repeat-dose toxicity studies in dogs, the primary nonrodent species used in regulatory safety assessment. However, their relevance to human safety is often uncertain. These findings may reflect local irritation, exaggerated pharmacology, off-target effects, formulation-related factors, or species-specific sensitivities. Determining their translational significance is critical for informing development decisions, including whether to progress a compound, modify formulation or dosing, or implement clinical monitoring strategies.

This course uses this translational challenge as a unifying framework to demonstrate how AI/ML approaches can be applied to integrate heterogeneous data sources, extract information from unstructured text, and support more consistent and transparent decision-making. The course is organized into four sequential sessions, each focusing on a distinct class of AI/ML methods. All sessions are anchored to a shared case study involving a hypothetical compound (“Compound X”) associated with GI toxicity findings in dog repeat-dose studies. This structure allows participants to directly compare how different analytical approaches influence interpretation and decision-making.

A defining feature of this course is its hands-on, interactive format. Each session includes a practical component in which participants will work in real time with pre-configured Jupyter notebooks and curated mock datasets that reflect realistic nonclinical and translational data. The notebooks will be fully set up in advance and designed for accessibility, requiring no prior programming experience. Participants will be guided through step-by-step workflows, including model development, data extraction, integration, and scenario analysis.

The first session introduces interpretable machine learning approaches for early signal detection and triage. Participants will explore how structured datasets derived from nonclinical studies can be used to train models that estimate the likelihood that dog GI findings are relevant to humans. Methods such as logistic regression and tree-based models will be used to illustrate how predictive performance can be balanced with interpretability.

The second session addresses the challenge of unstructured toxicology data. Toxicology reports and pathology narratives contain detailed observations that are not consistently captured in structured datasets. This session demonstrates how natural language processing (NLP) approaches can be used to extract and standardize this information into features suitable for analysis.

The third session focuses on multimodal data integration. Participants will explore how AI/ML approaches can combine diverse data types—including nonclinical findings, pharmacokinetics, and biological context—into a unified analytical framework to improve interpretation of toxicity signals.

The final session introduces causal inference and counterfactual analysis to support decision-making. Participants will evaluate how changes in exposure, formulation, or dosing strategy may influence GI toxicity risk using scenario-based approaches.

Throughout the course, emphasis is placed on practical application, transparency, and critical evaluation. Participants will consider both the capabilities and limitations of AI/ML approaches and their appropriate use in toxicology.

Chair(s):
Connie Mitchell, HESI; and Kamin Johnson, Corteva Agriscience.

Traditionally in toxicology, whole animal studies are designed to identify a dose at which no adverse effects are likely to occur for humans or species of concern. These reference doses are extrapolated from no-adverse-effect levels or benchmark doses (BMDs) that are used as a point of departure (POD). Transcriptomic dose-response modeling is a promising method that can produce PODs from gene expression profiling. A transcriptomic POD (tPOD) represents the lowest dose at which there is a detectable concerted change in gene expression in response to chemical exposure. At exposure levels below this tPOD, the likelihood of experiencing adverse toxicological effects is minimal. The tPOD is thought to be highly predictive of potential adverse toxicological effects that can be induced by a chemical stressor, making it a potentially valuable tool in risk assessment. Regulatory bodies are starting to consider the use of tPOD frameworks, like the US EPA’s Transcriptomics Assessment Product (ETAP) and Health Canada’s exploration of the use of tPODs with in vitro approaches.

In this course, participants will learn the scientific basis of why tPODs are appropriate for use in chemical risk assessment, including the biological rationale that coordinated changes in gene expression represent early indicators of cellular perturbation that precede overt adverse outcomes. These transcriptomic responses reflect the activation of stress response pathways, adaptive mechanisms, and toxicity pathways that are well established in mechanistic toxicology. Because these changes occur at doses lower than those causing apical effects, tPODs provide a sensitive and protective point of departure that aligns with the goal of identifying thresholds for potential harm. Participants will also learn how tPODs fit into current regulatory paradigms and how they compare to traditional apical PODs. The course will cover the key steps required to calculate a tPOD, including study design considerations such as dose spacing and replicate number, upstream data processing steps following sequencing such as outlier detection and removal, and downstream analytical decisions related to transcriptomic data filtering and model fitting. Finally, approaches for deriving a transcriptome-wide tPOD will be discussed, including comparisons between distribution-based and gene set–based methods, along with practical considerations for interpretation and application in risk assessment contexts.

This course is aimed at a wide audience, from those who have heard of a tPOD but are unfamiliar with the steps needed to calculate them, to those who are seasoned bioinformaticians but want to learn about current best practices based on recent research. For attendees interested in performing tPOD estimation themselves, BMDExpress files containing a completed analysis will be made available before the meeting. Some lectures will use this file to contextualize analysis settings as they appear in the software. However, attendees will not be required to perform a live analysis in BMDExpress during or prior to the course. The course is structured progressively, beginning with foundational concepts and regulatory context before moving into more technical bioinformatic and analytical considerations, ensuring accessibility for attendees with varying levels of transcriptomics expertise. Each talk will introduce a topic, showcase studies with publicly available data, and provide recommendations on process or practices to avoid.