The second session of the 2nd Joint BSTP/ESTP Toxicologic Pathology Congress, Manchester, UK, September 23-26, 2025, entitled “New Approach Methodologies for Carcinogenicity Evaluation,” was dedicated to innovative strategies for assessing carcinogenic risks in various substances, particularly in drug development and agrochemicals. The two-year rodent cancer bioassay in two species (generally rats and mice) is currently the standard method for assessing the carcinogenic potential of agrochemicals for humans. However, this method has some weaknesses and is subject to ethical and scientific debate. Attempts to waive those studies have been proposed, but more relevant methods using less or preferably no animals are still being sought. The session featured five key presentations that explored cutting-edge methodologies aimed at improving the accuracy and reliability of carcinogenicity predictions. Below is a summary of each talk presented.

This report summarises the main findings and discussion from the European Commission (EC) workshop on "The Roadmap Towards Phasing Out Animal Testing for Chemical Safety Assessments" which was held in Brussels on 11-12 December 2023. The aim of the workshop was to gather ideas and opinions from individuals, organisations and institutions, and to discuss potential approaches for incorporating non-animal methods into chemical legislation with all interested stakeholders. The roadmap will be an EC policy document which will outline milestones and specific actions, addressing all relevant pieces of chemical legislation relating to safety assessment. It intends to describe the necessary steps to replace animal testing in pieces of legislation where it is currently required for chemical safety assessments. The roadmap will outline the path to expand and accelerate the development, validation and implementation of non-animal methods as well as means to facilitate their uptake across legislation. The workshop included presentations from a wide variety of stakeholders. The contributors provided examples of how non-animal methods could be applied to replace, reduce or refine animal testing in the assessment of human health and environmental effects. Furthermore, possible guiding principles for establishing a Next-Generation Risk Assessment (NGRA) in European chemicals legislation were also discussed. The workshop provided the basis for further discussion and for structuring the roadmap work.

The European Partnership for Alternative Approaches to Animal Testing (EPAA) held the "New Approach Methodologies (NAMs) User Forum Kick-Off Workshop", at the European Chemicals Agency (ECHA), Helsinki, Finland on 7-8 December 2023. The aim of the User Forum was to gain insight into the regulatory use of NAMs, with a particular reference to Next Generation Risk Assessment (NGRA), for chemical safety assessment. To achieve this, presentations summarised the learnings and experiences of previous EPAA Skin Sensitisation User Forums as well as that of the European Commission's Scientific Committee on Consumer Safety (SCCS). The findings of five case studies were summarised that illustrated the use of NAMs. The presentations and subsequent discussions allowed for learnings and insights to be compiled from all stakeholders with regard to the use of NAMs. Recommendations for the regulatory use of NAMs in NGRA were made, namely for exposure assessment; hazard identification; using tiered and targeted testing strategies; performing risk assessment using NAM data; the practical implementation of NAMs; the use of -omics technologies; and the needs for capacity building and training. The EPAA User Forum provided an open platform for safety assessors to share learnings and experiences. Recommendations for the format and topics of future EPAA User Forums were also made.

By grouping structurally similar chemicals, toxicity endpoints from data-rich substances can be read across to data-poor substances, supporting environmental and human health risk assessment without animal testing. However, structural similarity alone is insufficient, and additional supporting data can strengthen a grouping justification. This study aimed to demonstrate how multi-omics bioactivity data can increase confidence in a grouping hypothesis, where the bioactivity profiles can reflect a chemical's mode(s) of action. We investigated three structurally similar phthalates and three uncouplers of oxidative phosphorylation, applying structure-based grouping approaches and short-term exposures of the ecotoxicological test species Daphnia magna to generate multi-omics data. Bioactivity similarities between the 'omics responses to chemical exposure were assessed using t-statistics comparing treated samples to controls and visualised using hierarchical cluster analysis. Conventional structure-based grouping did not assign the phthalates and uncouplers into two anticipated categories, with the structurally more diverse uncouplers often assigned into multiple groups. Following bioactivity thresholding, which removed one uncoupler as it induced minimal molecular responses, bioactivity profile-based grouping of the remaining five substances correctly separated them into two chemical classes with high replicability confidence. However, a plausible toxicological interpretation of the reduced set of functionally annotated molecular features driving the grouping was attempted, although of limited success. This study demonstrates how multi-omics bioactivity profiles can increase confidence in chemical grouping and investigates a potential strategy for plausibly interpreting 'omics data.

(Quantitative) structure-activity relationships ((Q)SARs) are widely used in chemical safety assessment to predict toxicological effects. Many thousands of (Q)SAR models have been developed and published, however, few are easily available to use. This investigation has applied previously developed Findability, Accessibility, Interoperability, and Reuse (FAIR) Principles for in silico models to six published, different, machine learning (ML) (Q)SARs for the same toxicity dataset (inhibition of growth to Tetrahymena pyriformis). The majority of principles were met, however, there are still gaps in making (Q)SARs FAIR. This study has enabled insights into, and recommendations for, the FAIRification of (Q)SARs including areas where more work and effort may be required. For instance, there is still a need for (Q)SARs to be associated with a unique identifier and full data / metadata for toxicological activity or endpoints, molecular properties and descriptors, as well as model description to be provided in a standardised manner. A number of solutions to the challenges were identified, such as building on the QSAR Model Reporting Format (QMRF) and the application of QSAR Assessment Framework (QAF). This study also demonstrated that resources such as the QSAR Databank (QsarDB, www.qsardb.org) are valuable in storing ML QSARs in a searchable database and also provide a Digital Object Identifier (DOI). Many activities related to FAIR are currently underway and (Q)SAR modellers should be encouraged to utilise these to move towards the easier access and use of models. Enabling FAIR computational toxicology models will support the overall progress towards animal free chemical safety assessment.

A broad range of computational models is available for animal-free chemical safety assessment. The models are used to predict a variety of endpoints, including adverse effects or apical endpoints, tox­icokinetic properties, and exposure, often from chemical structure or in vitro inputs alone. To support their wider use, such models need to be findable, accessible, interoperable, and reusable (FAIR). This study has reevaluated the existing FAIR principles applied to quantitative structure-activity rela­tionships (QSARs) in order to adapt these principles to a wider range of computational models. Despite the breadth and variety of approaches, many computational models comprise common components including the training series, information about the modelling engine, and the model itself. As a result, a refined set of four FAIR Lite principles is proposed based on the methodological foundations of computational toxicology which are unambiguously understood by practitioners such as developers and end-users. To this end, it is proposed that to comply with the original FAIR prin­ciples, a computational toxicology model should be associated with (i) a globally unique identifier for model citation; (ii) the capture and curation of the model; (iii) the metadata for the dependent and independent variables and, where possible, data; and (iv) storage in a searchable and inter­operable platform. The FAIR Lite principles are mapped onto the original FAIR principles applied to QSARs, thereby demonstrating that a simpler checklist approach covers all aspects.

The microcystins (MCs) are a family of cyclic oligopeptides toxins expressed in at least 30 cyanobacterial species and are liable to pose significant hazard to human health due to hepatotoxicity. Microcystin-leucine arginine (MC-LR) is the most extensively studied and toxic congener and classified as possibly carcinogenic to humans based on tumor promotion activity in the liver. Given the substantial toxicity data gaps for the MCs, read-across was assessed to evaluate the tumor promotion effects of a series of data-poor MC congeners based on in vivo information for MC-LR as the source molecule. Lines of evidence from in silico estimates of structural similarity, physico-chemical properties, hepatotoxicity, genotoxic and carcinogenicity were compiled to support the filling of data gaps. Uncertainties were evaluated according to scenario 4 of the European Chemicals Agency's (ECHA's) Read-Across Assessment Framework (RAAF). The read-across process followed a previously proposed harmonized framework to apply the common principles together with information from new approach methodologies (NAMs). Lines of evidence were consistent across the MC congeners and the uncertainties were found to be acceptable for data gap filling. Read-across strategies, with known caveats and restrictions, were shown to be applicable for large, complex molecules such as the MCs.

Quantitative structure-activity relationships (QSARs) are invaluable computational tools for the prediction of the biological effects and physico-chemical properties of molecules. For chemical safety assessment they are used frequently to make predictions of toxic or adverse effects, as well as other activities related to toxicokinetics. QSARs and their predictions can be assessed against a number of criteria for their potential use as surrogates for animal, or other, tests. A recent exercise by the Division of Genetics and Mutagenesis, National Institute of Health Sciences, Japan, assessed QSARs to predict the outcome of the Ames test. The predictive performance of models was scrutinised with full disclosure of results. The authors of this publication developed one such model, which had disappointing performance in this predictive exercise. In order to understand why the QSAR had poor performance metrics, this paper reflects on factors that affect a QSAR model. There is no one reason for poor performance of a QSAR model, rather it is likely to be a combination of factors. Reasons for poor performance included inadequate consideration of the underlying data quality, consistency and relevance; lack of appropriate descriptors relating to the endpoint and mechanism of action; not selecting a model correctly in terms of its structure (i.e., complexity) and number of descriptors; not addressing metabolism adequately in the modelling process; ill-defined assessment of the uncertainties within a model; and not ensuring predictions are within the applicability domain of the model. Whilst this paper draws on examples for the prediction of mutagenicity, the findings are applicable to all toxicological activities and physico-chemical properties.

Improving regulatory confidence and acceptance of in silico toxicology methods for chemical risk assessment requires assessment of associated uncertainties. Therefore, there is a need to identify and systematically categorize sources of uncertainty relevant to the methods and their predictions. In the present study, we analyzed studies that have characterized sources of uncertainty across commonly applied in silico toxicology methods. Our study reveals variations in the kind and number of uncertainty sources these studies cover. Additionally, the studies use different terminologies to describe similar sources of uncertainty; consequently, a majority of the sources considerably overlap. Building on an existing framework, we developed a new uncertainty categorization framework that systematically consolidates and categorizes the different uncertainty sources described in the analyzed studies. We then illustrate the importance of the developed framework through a case study involving QSAR prediction of the toxicity of five compounds, as well as compare it with the QSAR Assessment Framework (QAF). The framework can provide a structured (and potentially more transparent) understanding of where the uncertainties reside within in silico toxicology models and model predictions, thus promoting critical reflection on appropriate strategies to address the uncertainties.

Adverse outcome pathways (AOPs) were introduced in modern toxicology to provide evidence-based representations of the events and processes involved in the progression of toxicological effects across varying levels of the biological organisation to better facilitate the safety assessment of chemicals. AOPs offer an opportunity to address knowledge gaps and help to identify novel therapeutic targets. They also aid in the selection and development of existing and new in vitro and in silico test methods for hazard identification and risk assessment of chemical compounds. However, many toxicological processes are too intricate to be captured in a single, linear AOP. As a result, AOP networks have been developed to aid in the comprehension and placement of associated events underlying the emergence of related forms of toxicity—where complex exposure scenarios and interactions may influence the ultimate adverse outcome. This study utilised established criteria to develop an AOP network that connects thirteen individual AOPs associated with nephrotoxicity (as sourced from the AOP-Wiki) to identify several key events (KEs) linked to various adverse outcomes, including kidney failure and chronic kidney disease. Analysis of the modelled AOP network and its topological features determined mitochondrial dysfunction, oxidative stress, and tubular necrosis to be the most connected and central KEs. These KEs can provide a logical foundation for guiding the selection and creation of in vitro assays and in silico tools to substitute for animal-based in vivo experiments in the prediction and assessment of chemical-induced nephrotoxicity in human health.

This article aims to provide a comprehensive critical, yet readable, review of general interest to the chemistry community on molecular similarity as applied to chemical informatics and predictive modeling with a special focus on read-across (RA) and read-across structure-activity relationships (RASAR). Molecular similarity-based computational tools, such as quantitative structure-activity relationships (QSARs) and RA, are routinely used to fill the data gaps for a wide range of properties including toxicity endpoints for regulatory purposes. This review will explore the background of RA starting from how structural information has been used through to how other similarity contexts such as physicochemical, absorption, distribution, metabolism, and elimination (ADME) properties, and biological aspects are being characterized. More recent developments of RA's integration with QSAR have resulted in the emergence of novel models such as ToxRead, generalized read-across (GenRA), and quantitative RASAR (q-RASAR). Conventional QSAR techniques have been excluded from this review except where necessary for context.

In dietary risk assessment of plant protection products, residues of active ingredients and their metabolites need to be evaluated for their genotoxic potential. The European Food Safety Authority recommend a tiered approach focussing assessment and testing on classes of similar chemicals. To characterise similarity, in terms of metabolism, a metabolic similarity profiling scheme has been developed from an analysis of 46 chemicals of strobilurin fungicides and their metabolites for which either Ames, chromosomal aberration or micronucleus test results are publicly available. This profiling scheme consists of a set of ten sub-structures, each linked to a key metabolic transformation present in the strobilurin metabolic space. This metabolic similarity profiling scheme was combined with covalent chemistry profiling and physico-chemistry properties to develop chemical categories suitable for chemical prioritisation via read-across. The method is a robust and reproducible approach to such read-across predictions, with the potential to reduce unnecessary testing. The key challenge in the approach was identified as being the need for metabolism data and individual groups of plant protection products as the basis for the development of such profiling schemes.

Although uncertainties expressed in texts within QSAR studies can guide quantitative uncertainty estimations, they are often overlooked during uncertainty analysis. Using neurotoxicity as an example, this study developed a method to support analysis of implicitly and explicitly expressed uncertainties in QSAR modeling studies. Text content analysis was employed to identify implicit and explicit uncertainty indicators, whereafter uncertainties within the indicator-containing sentences were identified and systematically categorized according to 20 uncertainty sources. Our results show that implicit uncertainty was more frequent within most uncertainty sources (13/20), while explicit uncertainty was more frequent in only three sources, indicating that uncertainty is predominantly expressed implicitly in the field. The most highly cited sources included Mechanistic plausibility, Model relevance and Model performance, suggesting they constitute sources of most concern. The fact that other sources like Data balance were not mentioned, although it is recognized in the broader QSAR literature as an area of concern, demonstrates that the output from the type of analysis conducted here must be interpreted in the context of the broader QSAR literature before conclusions are drawn. Overall, the method established here can be applied in other QSAR modeling contexts and ultimately guide efforts targeted towards addressing the identified uncertainty sources.

Historically, regulatory decisions on the safety of chemicals to both humans and the environment have relied primarily on the availability of in vivo toxicity data to inform hazard and ultimately risk assessment. However, increasing recognition of the benefits of more mechanistically based scientific understanding, together with changing ethical and societal concerns, are driving the development of new approach methodologies (NAMs) that can support robust safety decision-making without animal testing. Grouping and read-across (G/RAx) is one of the most commonly used alternative approaches to animal testing in chemical risk assessment for filling data gaps with existing in vivo toxicity data (European Chemicals Agency [ECHA], n.d.; Organisation for Economic Co-operation and Development [OECD], 2017a). As such, it exemplifies the efficient use of existing data and in some cases new nonanimal data. For example, under REACH (Registration, Evaluation, Authorisation and Restriction of Chemicals regulation) Annex XI, information from one or more analogous (or "source") chemicals can be used to predict missing endpoint data for one or more "target" chemicals (European Commission, 2006). With approximately 100,000 chemicals listed on the European inventory (ECHA, 2023) and approximately 85,000 chemicals listed in the US Environmental Protection Agency's (USEPA's) Toxic Substances Control Act (TSCA) inventory (2024a), the use of G/RAx (described as chemical "categories" under the TSCA; USEPA, 2010) is becoming an increasingly viewed option for addressing regulatory requirements for filling data gaps in chemical safety dossiers for human health and environmental endpoints. Furthermore, grouping of chemicals can facilitate other hazard-assessment practices, for example, the harmonized classification of multiple substances within a group in accordance with the classification, labeling, and packaging regulation (Swedish Chemicals Agency, 2020). There are numerous approaches for defining groups of chemicals, most often based on chemical similarity (Patlewicz et al., 2018). Notable examples in a regulatory context include the approach documented in the ECHA Read-Across Assessment Framework (RAAF; ECHA, 2017), supporting REACH, and within the TSCA (USEPA, 2010). These existing schemes are traditionally and primarily based on firstly grouping "source" and "target" chemicals into categories based on structural and other physicochemical parameters and, secondly, reading across existing toxicity data (i.e., an apical endpoint) from one or more "source" chemical(s) to predict the toxicity of one or more "target" chemical(s). However, most grouping dossiers still fail to incorporate and utilize absorption, distribution, metabolism, and excretion (ADME)/toxicokinetic and toxicodynamic similarities, with the strong reliance on structure-based similarity often leading to a rejection of the proposed read-across arguments, potentially resulting in regulatory noncompliance. For example, solely relying on structural similarity as the justification for a read-across introduces the potential to misevaluate the hazard of the target because structural similarity does not strongly infer equivalent levels of toxicity. This has prompted new efforts, such as the National Institute of Environmental Health Sciences workshop on clustering and classification (2022), to increase the confidence and consistency of chemical grouping by integrating molecular responses, and ideally a mechanistic understanding, into this process (Escher et al., 2019; Pestana et al., 2021). While NAMs span a wide range of approaches from in vitro testing and novel bioanalytical assays to in silico methods, in our study we focus on the application of omics technologies to generate molecular data that can be used to quantitatively determine group membership, thereby offering a solution to a significant limitation of conventional structure-based G/RAx approaches. This approach to forming chemical groups involves quantitatively comparing "profiles" of biological response data, derived from omics technologies such as transcriptomics (measuring gene expression) or metabolomics (measuring downstream metabolic biochemistry), and in concept is not unlike the widely used approaches for comparing structural fingerprints such as Tanimoto similarity (Sperber et al., 2019). Furthermore, with metabolomics possessing the capability to measure substance metabolism, there exists the potential to utilize both ADME/toxicokinetic and toxicodynamic similarities to build reliable groups from this data type. However, progress incorporating omics data into G/RAx has been hampered by a range of factors, including siloing of new scientific developments from regulatory science, to more specific issues such as a lack of standardized assays, reporting templates, and well-constructed case studies. To introduce G/RAx to nonexperts, its importance, its terminology, the legislation through which it operates, and, most importantly, its current limitations. To introduce omics technologies to regulatory scientists, as applied in the context of G/RAx, explaining the value of applying these molecular assays to quantitatively group chemicals. To introduce the reader to some grouping case studies that use omics data, thereby increasing awareness of how these approaches have been used to group chemicals. To describe some challenges to advancing the incorporation of omics data into chemical grouping and identify next steps toward accelerating this implementation. To achieve our goal of introducing chemical grouping based on omics data across a range of contexts of use and regulatory jurisdictions, this article is necessarily generalized in places. Read-across is routinely used to predict an apical endpoint of a chemical by interpolating or extrapolating the toxicity data of analogous chemicals that are similar in some manner, for example, chemical structure, shared metabolism, and/or mode of action (MoA; which defines a functional cellular change), thereby avoiding further testing. A schematic summarizing the concepts of conventional G/RAx is shown in Figure 1, and relevant terminology is introduced in Textbox 1. It is based on an assumption that a physicochemical, (eco)toxicological, or environmental fate property of a "target" compound can be inferred from test data for the same property of similar "source" compounds (OECD, 2017a). There are two distinct approaches to read-across: an analogue approach describes read-across from a single or very small number of source chemicals to a target chemical, whereas a category approach is used when data from a larger group of source chemicals are read-across to the target(s). Read-across predictions are endpoint-specific; for example, in quantitative read-across a known value(s) of a single endpoint for a source chemical(s) is used to infer a quantitative value of the same endpoint for the target chemical. Grouping—process of forming groups (or categories) of chemicals that have similar (or follow a regular pattern of) physicochemical, (eco)toxicological, and/or toxicokinetic properties. Read-across—alternative method for obtaining toxicity data (i.e., endpoint information) for one chemical—the target—by using data from the same endpoint from another chemical(s)—the source chemical(s), also referred to as an analogue—where the source and target chemicals lie within the same group. Endpoint—definition depends on the context of use. In REACH information requirements, endpoints are described either as a toxicological property (e.g., skin irritation, long-term toxicity to aquatic organisms) or as a type of study (e.g., carcinogenicity study, Daphnia chronic assay). Grouping hypothesis—description of the proposed membership of one (or more) chemical groups, based on the similarities of structural (or other physicochemical), (eco)toxicological, and/or toxicokinetic properties. Category justification—reasoning and associated evidence to verify the scientific validity of the grouping hypothesis for three or more chemicals. For the specific case of a single source and single target substance, this reasoning would be termed an analogue justification. Bridging studies—comparable studies on the source and target chemicals that allow a direct side-by-side comparison of the chemicals for a particular toxicological property (OECD, 2017a). Existing global regulations endorse the use of grouping as the basis for reading across existing toxicity data to fill data gaps for industrial chemicals. For example, in the United States, the TSCA requires consideration of chemical grouping, stating in Section 4(h) that as part of reducing and replacing vertebrate animal testing of chemicals it encourages "the grouping of 2 or more chemical substances into scientifically appropriate categories in cases in which testing of a chemical substance would provide scientifically valid and useful information on the chemical substances in the category" (USEPA, 2018). Under the TSCA, the USEPA already routinely uses chemical grouping techniques such as the classification applied in the Ecological Structure–Activity Relationship (ECOSAR) software, new chemical categories, and analogue identification to fill data gaps in the assessment of new chemical submissions. The USEPA also considers analogues for data gap filling of existing chemicals and in special cases, such as per- and polyfluoroalkyl substances, has developed rules to group chemicals to direct national testing strategies. Associated with these practices, the USEPA is drafting guidance for how to select and use analogue data in ecotoxicology in place of animal data. In parallel, it is noteworthy that TSCA Section 4(h) states that information from "high throughput screening methods and the prediction models of those methods" should be considered "prior to making a request or adopting a requirement for testing using vertebrate animals," providing a legal route toward utilizing molecular (including omics) data for chemical grouping. Complementing these regulations, extensive international guidance also exists for describing how to assess the hazards of related chemicals as a group, rather than as individual chemicals. Foremost is guidance published by the Chemical Safety Programme of the OECD, which assists member countries in their efforts to protect human health and the environment from hazardous chemicals using the best available science. The OECD is also committed to promoting alternatives to animal testing when suitable methods for evaluating chemical safety can be demonstrated. Of particular note to our study is the OECD Guidance on Grouping of Chemicals (Series on Testing & Assessment No. 194; 2017a), which focuses on considering the hazards of chemicals as a group or category. First published in 2007, it was updated in 2014 to include sections on analogue and category approaches, quantitative and qualitative read-across, justifying read-across, and using "bioprofiling" results (that include molecular data) for grouping chemicals. A new edition of this guidance document is now being prepared, including more extensive guidance on the use of omics data for chemical grouping. A further example of guidance related to chemical grouping, in this case specifically grouping and read-across, is the RAAF published by the ECHA (2017). This document provides both a framework and guidance for describing how G/RAx should be used and presented in a registration dossier in the context of meeting the REACH information requirements. It facilitates the consistent evaluation of the elements within a read-across case, including the grouping hypothesis and category (or analogue) justification. Case studies represent a further important contributor to defining future regulatory landscapes by helping to establish common and best practices for the use of novel methods for assessing chemicals, including as groups. For example, case studies form the Accelerating the Pace of Chemical Risk Assessment initiative, which is an international government-to-government activity whose aim is to promote collaboration and dialogue on the scientific and regulatory needs for the application and acceptance of NAMs in regulatory decision making (https://apcra.net/). These include a case study applying multi-omics to chemical grouping (Gruszczynska et al., 2024). The OECD Integrated Approaches for Testing and Assessment (IATA) case studies project aims to provide a forum to exchange information and grow confidence in the application of NAMs to assess chemical hazards in specific regulatory contexts (OECD, n.d.). A total of 34 IATA case studies have been reviewed, discussed, and published on the IATA OECD website (https://www.oecd.org/chemicalsafety/risk-assessment/iata/). Of these, 20 case studies have included aspects related to grouping and read-across, of which two have used transcriptomics data (OECD, 2017b; OECD, 2020). Structured reporting templates are critical tools for ensuring that standardized information is consistently described, facilitating both the evaluation of chemical toxicity and ecotoxicity data by regulators and data sharing. They also allow end users to assess if the approach may be suitable for other geographical regions, chemical sectors, or regulatory contexts and most importantly allow regulators to become familiar with data derived from NAMs. Reporting formats for documenting conventional chemical grouping and read-across—including the context of use, target chemical(s)/category definition, endpoint, grouping hypothesis, and justification for filling data gaps—are available (see, e.g., Chapter 7 in OECD, 2017a). An OECD project has developed the Omics Reporting Framework (OORF) for describing the acquisition, processing, and analysis of omics data for a range of applications (OECD, 2023). The OORF is currently being extended to include a reporting template for the specific application of molecular (omics) data to chemical grouping. The OECD states that "the most compelling evidence in support of a read-across hypothesis is information on a common mode of action of the substances and a mechanistic rationale for their common biological behaviour" (OECD, 2017a). Conventional G/RAx approaches are based on the hypothesis that structurally similar chemicals elicit similar biological responses. Yet there are many groups of chemicals for which this assumed relationship between structure and function does not hold. For example, thalidomide exists as two enantiomers that produce distinct biological responses: one is a sedative, and the other causes fetal malformations. Such "activity cliffs"—where structurally similar compounds have different MoA and/or potencies—demonstrate that structural similarity does not always lead to similar biological responses. By contrast, fentanyl and morphine are structurally dissimilar compounds, yet both share the same MoA, thus producing similar effects. The fact that structure does not always reliably predict function was well articulated by Wallqvist et al. (2006), who reported that "the connection between structure and biological response is not symmetric, with biological response better at predicting chemical structure than vice versa. Structurally and functionally similar compounds can have distinguishable biological responses reflecting different mechanisms of action." Furthermore, such differences in the toxicodynamic properties (i.e., MoA) of analogous compounds may also manifest as different acute and/or chronic toxicity outcomes. This limitation is acknowledged in ECHA's RAAF, which states "structural similarity alone is not sufficient to justify the possibility to predict properties of the target substance by read-across." It is therefore no surprise that Schultz et al. (2019) identified one of the main sources of uncertainty in read-across as the category justification (i.e., justifying group membership). If the source chemical(s) is not sufficiently similar to the target chemical(s) in terms of structure, to support similarity in their MoA, read-across will not be justifiable. For this reason, uncertainty surrounding the category justification must be addressed because it weakens confidence in the grouping, discussed further below. Under REACH, if read-across within a registration dossier does not provide sufficient evidence for a robust grouping of source(s) and target(s) chemicals, the dossier could be deemed noncompliant. Specifically, the ECHA has reported that G/RAx studies can be rejected in the absence of supporting data to substantiate the grouping hypothesis, for example, lack of knowledge of MoA and/or bridging studies between the source and target chemicals (ECHA, 2020). A second cause of registration dossiers that incorporate G/RAx being rejected is when the rationale for the read-across is missing or weak; for example, no explanation is provided linking structural similarity with the predicted toxicological endpoint. Another common problem when forming chemical groups based on ECOSAR and other quantitative structure–activity relationship ([Q]SAR) predictions is when many chemicals fall into multiple groups or into none at all. This is a particular problem where there are no or too few measured data to enable any benchmarking of the predicted values. There is clearly a need to improve our confidence in the formation of chemical groups or categories such as those efforts undertaken by the Board of Scientific Counselors in updating the TSCA (USEPA, 2022) and to quantify that level of confidence to facilitate the use of grouping in various risk-assessment contexts. The use of biological effects data as a basis for providing evidence to support group (or category) formation—via calculating the similarity of responses to chemical exposure—is widely viewed as a feasible solution for reducing uncertainty in grouping. However, we must ensure that the biological data will provide sufficient confidence for this application. It is important to emphasize that the generation of meaningful biological data requires a consideration of both the biological test system and the applied molecular assay. Several options are available for measuring molecular responses to exposure. For the generalized case of a data-poor target (i.e., limited experimental toxicity data), a biological effects comparison of source and target chemicals should examine a broad range of possible MoAs or mechanisms of action (MechoAs) to ensure that any significant toxicological effects are covered. In contrast, if the question is whether a target chemical behaves similarly to a group of chemicals with a relatively well-defined MoA or MechoA (which defines a specific target or pathway), a greater weighting could be placed on assessing the similarity of biological effects specific to that MechoA/MoA. Molecular measurements from omics technologies can be used to help address both of these situations, as illustrated in Figure 2, with introductory terminology presented in Textbox 2. For the former case, "untargeted" omics technologies measure a broad range of molecular responses (thus informing a broad range of MechoAs/MoAs); for the latter, a defined panel of "key event" molecular biomarkers could be measured that, for example, were derived from an adverse outcome pathway (AOP) describing the MechoA/MoA (https://aopwiki.org/). This first step in the application of omics technologies to chemical grouping is depicted in Figure 3 (right side). Ome—a broad collection of biomolecules, for example, genome (all genetic material), transcriptome (all gene transcripts), proteome (all proteins), and metabolome (all small-molecule metabolites). Omics—technologies that can be used to measure a broad range of molecular responses in the genome, transcriptome, proteome, or metabolome of a biological test system following chemical exposure. Transcriptomics—systematic study of expression of many genes in a cell, tissue, or organism, providing information on the molecular responses following chemical exposure. Metabolomics—systematic study of levels of many small-molecule metabolites and the biochemical processes that they are involved in, within a cell, tissue, or organism, providing information on downstream functional molecular responses to exposure. Targeted assay—measurement method that predefines the analytes, for example, to measure a specific MechoA/MoA (if few analytes are targeted) or multiple MechoAs/MoAs (if many analytes are targeted). Measurements using omics technologies can be targeted (see Table 1). Untargeted assay—measurement method that does not predefine the analytes. This approach attempts to measure the broadest range of molecular responses and hence gain insights into multiple MechoAs/MoAs simultaneously. Measurements using omics technologies can be untargeted (see Table 1). Key event molecular biomarker—a measurable marker that serves as an indicator of a specific MechoA/MoA, for example, derived from an AOP. Molecular effects data—a type of NAM data derived from either targeted or untargeted molecular assays that is used to calculate the bioactivity similarity of two or more chemicals and which can provide insights into MechoA(s)/MoA(s). Bioactivity similarity—a quantitative measure of similarity of two or more chemicals that is derived by comparing the molecular effects data from omics assays. It is important to highlight that measuring and using molecular data in regulatory toxicology are not new. Of the three measurement strategies for grouping, described in Table 1, the first—targeted molecular measurements—is already part of internationally accepted OECD test guidelines. The second strategy multiplexes (or parallelizes) many targeted molecular measurements, and the third approach additionally introduces untargeted measurements. Examples are provided in Table 1 to support the reader's understanding of these measurement strategies for a generalized MechoA/MoA assessment. Having generated either targeted or untargeted omics data in a relevant biological test system, the next step is to calculate the bioactivity similarity of the molecular responses to the source and target chemicals (see definitions in Textbox 2), in what can be referred to as a bridging study (Textbox 1). This typically involves applying multivariate statistical analyses and can enable assigning a quantitative level of confidence to the grouping hypothesis (Figure 3, right side). The third step is to interpret the molecular data in an attempt to identify the MechoA/MoA/AOP to build evidence for the grouping hypothesis; that is, if the omics data identify shared MechoA/MoA/AOP(s) that are perturbed as a result of exposure to a series of chemicals, then this provides a mechanistic justification to group those chemicals (Figure 3, right side). Minimally, evidence of shared molecular effects (without a clear understanding of mechanism) can be used to provide some justification for the chemical category. Here, resources that associate molecular changes with pathway perturbations, MechoA(s)/MoA(s)/AOP(s), and/or hazard are important parts of the toolkit. Many open-access and commercial resources exist, including the AOP Wiki, the Comparative Toxicogenomics Database, the S1500+ gene panel, the MTox700+ metabolite panel, BASF's MetaMapTox, and Qiagen's Ingenuity Pathway Analysis. By using omics data to substantiate a grouping hypothesis, the conventional G/RAx workflow (e.g., OECD, 2017a) need only be slightly altered to include this additional supporting evidence (Figure 3, left side). The initial steps—identifying the target compound and selecting source chemicals based on structural and physicochemical similarity—can remain largely the same. This satisfies current requirements for read-across to be based on structural similarity between source and target chemicals. The next step would typically be to refine the categories using mechanistic (MechoA/MoA/AOP) and/or endpoint-specific profilers (e.g., using the OECD [Q]SAR Toolbox). It is at this stage that the omics data can be introduced; that is, in addition to using these profilers, chemical categories are also refined based on bioactivity similarity using omics-derived molecular data. Source chemicals that are shown through a bridging study to not be biologically similar to the target can be excluded from the group. This approach should provide more confidence that the source and target chemicals share a common MechoA/MoA/AOP and thus elicit similar toxicological endpoints. Once the final list of source chemicals has been produced, the available toxicity data can be read across to fill the data gap for the target chemical. In effect, this is a weight-of-evidence approach that utilizes structural and bioactivity similarity to derive the chemical categories. In the future, if a structure-based chemical group is inadequate, it may be useful to create an alternative grouping hypothesis based solely on omics-derived molecular data. Such a change to regulatory practice would first require greater confidence in bioactivity profile–based grouping using omics data than exists today and would be best facilitated by the availability of a database of omics responses to enable comparison to data generated for the target substance. Several bioactivity profile–based grouping (and read-across) studies using omics data have been published, demonstrating the interest in evaluating this approach for supporting effective decision-making. We selected five studies and mapped them against a series of questions to highlight what each study sought to achieve, what approaches were used including whether any guidelines were followed, and to what extent uncertainties were considered to enable the reporting of the confidence of the grouping (Supporting Information, Table S1). Although such a mapping exercise is helpful for highlighting emerging "acceptable practice," in our study we focus on examining the benefits (this section) and challenges (next section) of incorporating omics data into chemical grouping (summarized in Textbox 3). Summary of top five benefits and challenges of incorporating omics data into chemical grouping. To provide targeted measurements of a broad range of MechoAs/MoAs simultaneously and an untargeted assessment of uncharacterized MechoAs/MoAs (dependent on the biological test system). To enable grouping by quantitatively comparing the similarities and differences of omics bioactivity profiles across chemicals, using statistically derived probability estimates, as a complement to structural similarity. In addition, the potential to provide mechanistic insights alongside statistically derived bioactivity-based grouping, together contributing to the category (or analogue) justification to support read-across to fill data gaps. Can be applied in high-throughput in vitro studies to screen and group the effects of multiple substances, enabling the triggering of higher-tier testing of a few sentinel substances. Can be used to group substances even when their structures are ill-defined, for example, UVCBs and polymers. A complex type of molecular data that requires specialist expertise for data analysis and interpretation. Currently, level(s) of bioactivity similarity has not been determined to assess "how similar is similar enough" to place two or more substances into the same group, nor have bioactivity thresholds been determined to delineate a molecular effect from no effect. It can be difficult to interpret the molecular changes and associate them to a MechoA/MoA. Currently, there has been only a limited assessment of the reliability of bioactivity profile–based grouping using omics data across laboratories, although there is ongoing work to address this challenge. There is currently no agreed "best practice" for grouping using omics data, although progress is underway at the OECD. The key benefit is that grouping using omics data does offer a solution to the well-recognized problem that grouping hypotheses typically do not include biological effects data; that is, they lack sufficient mechanistic underpinning. Furthermore, by measuring a broad swathe of biological response space (i.e., multiple MechoAs/MoAs/AOPs) simultaneously, and without bias, the untargeted omics measurements enable a data-driven assessment of not only which chemicals group together but also what molecular perturbations are driving that grouping. This knowledge could provide insights into hazards and, consequently, could aid the selection of further targeted assays. Further benefits of bioactivity profile–based grouping using omics data arise from the inherent capabilities of the technologies and computational approaches being applied. For example, metabolomics analyses can be conducted on biofluids, thereby enabling repeated measurements of some biological test systems, which could be valuable in cases where extensive metabolism or bioactivation occurs. Furthermore, the data analysis workflows are able to provide quantitative measures of similarity between two (or more) omics profiles, and confidence in the grouping can be based on statistical probabilities, facilitating robust decision-making (Gruszczynska et al., 2024). Although we are proposing that bioactivity profile–based grouping be considered as part of the weight of evidence toward a grouping hypothesis based on structural similarity, this assumes that structures can be defined for the test substances. However, there are many substances, for example, those of unknown or variable composition, complex reaction products and biological materials (UVCBs), that are poorly characterized in terms of component chemicals and their proportions, for which structure-based grouping is difficult, if not impossible. Applying an untargeted omics assay to an appropriate biological test system and calculating the bioactivity similarity from the biolo