Abstract: The European Union’s Chemicals Strategy for Sustainability (CSS), first published in October 2020, is supported by five objectives. One key goal, “Simplifying and consolidating the legal frameworks concerning chemicals in the EU“, relates to the One substance, one assessment (1S1A). Built on four pillars ―safety assessment initiation, allocation of tasks, data and methodologies―, the 1S1A envisages to improve scientific robustness, efficiency, transparency and coherence of chemicals’ risk assessment and regulation. EFSA is actively implementing the 1S1A concepts with several activities, including: ‘Early identification of cross-cutting substances’​​, ‘Piloting the 1S1A approach’, ‘Study to map Data Requirements (DR) and Risk Assessment Methodologies (RAM)’, ‘Contribution to the EU- Common Data Platform on Chemicals (EU-CDPC)’. The early identification of cross-cutting substances is being tested with those chemicals (on on-going or finished assessments) behind ECHA, EMA, the EC Scientific Committees and the food/feed assessments of EU Member States; the searching strategies and the application of 1S1A concepts (e.g. data sharing) are being examined. Whilst piloting the 1S1A approach, some misalignments in data requirements between chemicals regulatory framework were identified. That was the main driver for the study to map the DR and RAM across the various regulatory areas on chemicals in the EU, aiming to identify commonalities and differences across the operating frameworks. The establishment of the EU-CDPC containing chemical data is a major objective of the 1S1A. Several building blocks will compose the platform such as a repository of health-based-guidance values, a central database on regulatory submissions (IUCLID) and an occurrence data (IPCHEM). Legislation supporting the 1S1A ―focusing on the CLP revision, the re-attribution of tasks to EU Agencies, and the regulation on data― along with its likely impact to EFSA, will also be presented.   read more...   read less

Abstract: Novel Foods are defined as foods not consumed to a significant degree in the European Union before 15 May 1997. The market for Novel Foods has evolved significantly over the past 28 years, driven by scientific innovation and shifts in consumer demand. This project explored these evolving trends and their implications for regulatory assessment of product safety. A detailed analysis was conducted to identify changes in Novel Food applications submitted to the European Union and United Kingdom markets since 1997. Applications were categorised into five major intended uses, selected for their regulatory relevance and ability to capture diverse application types: nutritional enhancement, health claims, functional ingredients, traditional foods, or sustainability. The annual proportion of each category was calculated to identify emerging trends and areas of decline over time. Historically, Novel Food applications centred on health claims and consisted primarily of whole plant extracts for use as supplements. More recently, applications increasingly feature sustainability as a driver, such as cell-cultivated meat, insect protein, and precision-fermentation. Innovative Novel Food categories present new regulatory challenges, including significant shifts in the potential for allergenicity and toxicity, and the growing importance of ‘omic technologies. These new challenges may require additional data to meet safety requirements during a time when the burden for data generation, particularly the use of animal testing, is being reconsidered. Understanding shifts in Novel Food applications can help to guide the development and use of New Approach Methodologies (NAMs) by applicants. These tools can enhance both the quality and efficiency of safety assessments, allowing applicants to focus safety testing and data generation efforts where most needed.   read more...   read less

Abstract: Recent advances in artificial intelligence (AI), machine learning (ML), and deep learning (DL) are leading to a new era of data-driven innovation, significantly impacting the field of chemical risk assessment. These advanced computational techniques are increasingly being recognized as powerful tools for various applications in toxicology, including data retrieval (e.g. systematic reviews), predictive modeling (e.g. toxicokinetics, quantitative structure-activity relationships (QSARs), read-across), toxicity prediction, and evidence integration. To effectively and reliably integrate these tools into regulatory frameworks, it is essential that regulatory authorities, policymakers, and other stakeholders gain a deeper understanding of their nature, applicability, validity, and inherent limitations. It is equally crucial for computational scientists and AI developers to understand regulatory requirements and prerequisites. This knowledge enables them to develop methods that meet regulatory standards while effectively communicating both the advantages and limitations of these methods for regulatory applications. To maximize the benefits of AI in chemical safety assessment, a skilled and multidisciplinary task force is needed, fostering collaboration among regulators, toxicologists, computational scientists, and AI developers for responsible utilization of AI- driven methodologies. Within this framework, and as part of the Partnership for the Assessment of Risks from Chemicals (PARC) initiative, we have developed the project READYAI. We aim to address the key challenges associated with implementing AI in toxicology by developing a comprehensive scoring system for evaluating the development, training, and validation of AI-based tools applied in chemical risk assessment, enabling a more independent and objective assessment of the regulatory applicability of AI/ML/DL-based tools. Key concerns include ensuring the reliability of AI-generated predictions, mitigating biases in training data, improving algorithm interpretability, explainability, reproducibility & transparency, and addressing legal and ethical implications. By defining readiness criteria and adhering to best practice principles, READYAI seeks to facilitate regulatory acceptance of AI/ML/DL tools applied in chemical risk assessment.   read more...   read less

Abstract: Chemical risk assessment is undergoing a significant transformation driven by technological advancements that facilitate hazard and exposure characterization through high-throughput assays, reducing the need for animal models. This evolution has led to the development of the Next- Generation Risk Assessment (NGRA) framework, which combines bioactivity data with internal human exposure to establish protective thresholds for human health. The framework emphasizes the integration of various New Approach Methodologies (NAMs) to assess systemic toxicity. While the NGRA approach has been successfully applied in the cosmetic industry following the 2013 European ban on animal testing, its implementation in the food sector remains limited due to existing regulatory requirements. A NAMs toolbox, originally developed in other industry sectors, was adapted and applied to food-relevant chemicals. It incorporates methods such as cellular stress responses, in vitro pharmacological profiling, high-throughput transcriptomics and physiologically- based kinetic (PBK) modelling, to derive Bioactivity-Exposure Ratios (BER), demonstrating the potential of NGRA to inform safety decisions without animal testing. In parallel, a peer-review publication we co-authored under FoodDrinkEurope highlights the growing integration of NAMs in food safety assessments, with the European Food Safety Authority (EFSA) aiming to minimize animal testing by 2027. Despite this ambition, the use of NAMs in EU food evaluations remains limited due to existing regulations requiring animal studies. The review categorizes fourteen relevant NAMs and suggests amendments to the EU regulatory framework to enhance NAMs’ implementation in a 3R perspective. It emphasizes the need for revisions to existing guidance, advocating that animal testing should be a last resort and calling for research initiatives to build confidence in alternative methods. Together, these efforts aim to create a more ethical and efficient approach to food safety assessments, aligning with evolving scientific advancements and societal expectations.   read more...   read less

Abstract: The Collaboration to Harmonise the Assessment of Next Generation Evidence (CHANGE) focuses on system factors as an area of research that could be further developed and customized for the effective use of NAMs in regulatory toxicology. By “system factors”, we mean potentially modifiable aspects of the organisational environment in which NAM-based research is conducted, interpreted, and acted on. System factors, typically including technical and social factors (e.g., people, resources, culture, processes, policies, and legislation), may provide new opportunities for the effective implementation of NAMs in regulatory toxicology. CHANGE is an interdisciplinary collaboration including an international group of people in the regulatory toxicology system having different roles and focuses, different specialisations, working in different sectors with different types of chemical substances covered by different legislations. The project is based on involvement, networking, and dialogue. The goal of CHANGE is to design system-level interventions for bringing forward the date of effective use of NAMs. CHANGE follows a three-phase methodological approach “explore—reflect—design”:  Explore: Identify factors that might matter for the uptake and effective use of NAMs through assessing and describing mental models of the regulatory toxicology system by people operating within it.  Reflect: Identify the factors that actually do matter by challenging the mental models and hypotheses created in the explore phase.  Design: Develop interventions addressing the factors that matter. System diagrams will be used as a tool to visualise the processes and dynamic forces in the regulatory toxicology system and interactions between them, which will be used to identify opportunities for improving the effective use of NAMs. Building on previous and ongoing efforts and initiatives, having a holistic approach and focussing on system factors, CHANGE aims to be a catalyst for bringing forward the date of effective use of NAMs in the regulatory toxicology system.   read more...   read less

Abstract: The EU’s shift toward non-animal, mechanistic chemical safety assessment places New Approach Methodologies (NAMs) at the forefront of regulatory toxicology. Endocrine disruption (ED), especially via thyroid-related mechanisms, remains a scientifically and regulatory complex issue across multiple sectors, including agrochemicals, biocides, industrial chemicals (REACH), and pharmaceuticals Limitations of traditional in vivo testing, and species-specific differences make thyroid related assessments particularly complex. NAMs such as in vitro assays, in silico models, and omics approaches can provide mechanistic insights and enhance the evaluation of human relevance. However, their acceptance in regulatory decision-making is still limited. EFSA and ECHA promote weight-of-evidence and next-generation risk assessment (NGRA), and OECD test guidelines were recently updated to include relevant ED endpoints. Still, validated NAMs for ED assessments, particularly for thyroid disruption, are lacking and regulatory acceptance remains in its early stages. Efforts are underway to support the integration and validation of NAMs through EU initiatives such as EU-NETVAL, PARC, the EURION and ASPIS cluster. A coordinated strategy based on Adverse Outcome Pathways (AOPs), species comparison, and early dialogue is essential to foster regulatory confidence across sectors.   read more...   read less

Abstract: Many promising new approach methodologies (NAMs) resulted from huge investments in research but only a few are validated. Several are meanwhile widely established and exploited in industrial settings. Hence, regulators are increasingly faced with data from non-guideline studies demanding for a complex and time-consuming case-by-case evaluation. Importantly, for nanomaterials (NMs) most NAMs (with only a few notable exceptions) are not yet formally validated. While validation permits the use of a NAM for a wide range of regulatory applications a qualification is a process that facilitates the use of a NAM in a particular context (“context-of-use concept”). In fields with high innovation such as nano- and advanced materials, a qualification could be well suited to allow for faster regulatory application of non-guideline methods in a particular context in a timely manner, once they had been examined in accordance with predefined criteria. Within the EFSA-funded project NAMs4NANO we proposed a qualification system for NM safety assessments in the food and feed sector. We considered available information such as existing qualification systems operating in the context of Drug Development Tools (US FDA) and research and development into pharmaceuticals (EMA), published frameworks/ approaches for method readiness evaluation, guidance documents (e.g., EURL-ECVAM, OECD) and ongoing efforts (e.g., PARC) to propose a generic qualification framework. We also proposed readiness criteria as guidance covering different types of NAMs (in silico, in chemico, in vitro). We test the system and the criteria for selected NAMs in the context of five risk assessment case studies which are ongoing in our project. By no means method qualification should be regarded as an alternative to validation. However, a qualification system can significantly accelerate the identification of potentially promising NAMs and endorse those which are demonstrated to be scientifically valid for regulatory application for a specific intended use in a well-defined context.   read more...   read less

Abstract: Switzerland’s regulatory framework for food supplements and cosmetic products reflects its unique multilingual and federal structure, offering a robust model for consumer protection and market integrity. Unlike many EU member states, Swiss law requires product labelling in at least one of the country’s official languages - German, French or Italian - depending on the region of sale. This multilingual requirement enhances transparency and accessibility for consumers, while also presenting operational considerations for international manufacturers and distributors. Our study explores the implications of Switzerland’s labelling obligations and its alignment with EU regulations, particularly in the domains of food supplements and cosmetics. While Swiss legislation broadly mirrors EU food law, it also introduces distinct safeguards - such as detailed origin declarations and stricter controls on certain substances - administered by the Federal Food Safety and Veterinary Office (FSVO). These measures underline Switzerland’s commitment to public health and product safety. We analyse how these regulatory features influence product formulation, packaging, and market entry strategies, and highlight the importance of proactive compliance and self-supervision by manufacturers, especially given the absence of pre-market authorization for most food supplements. Our case examples illustrate how multilingual labelling, when not carefully managed, can lead to inconsistencies - yet also demonstrate how FSVO guidance and collaborative approach help mitigate such risks. By positioning Switzerland as a case study, our work contributes to the broader discourse on regulatory harmonization, consumer protection and multilingual compliance. It emphasizes the value of clear regulatory guidance and international cooperation in fostering safe, transparent and equitable access to food supplements and cosmetic products across diverse linguistic and legal environments.   read more...   read less

Abstract: In the post-COVID-19 era, Armenia confronts a dual epidemiological burden: the persistence of long COVID (LC) and a parallel surge in noncommunicable diseases (NCDs), especially among middle-aged and older adults. These trends are compounded by immunosenescence and widespread misuse of antibiotics and corticosteroids, which disrupt gut microbiota and contribute to chronic low-grade inflammation ("inflammaging"), further exacerbating LC severity. National survey data reveal significant post-COVID sequelae, with high rates of fatigue, musculoskeletal pain, anosmia/ageusia, cognitive impairment, and self-medication. Using WHO-endorsed methods, the national burden of LC was estimated at approximately 215,934 Disability-Adjusted Life Years (DALYs) and 20,104 Quality-Adjusted Life Years (QALYs) lost. To address these challenges, a digital health strategy integrating Artificial Intelligence (AI) and an interoperable platform was developed to support early triage and personalized case management. Central to this approach is a composite scoring methodology, leveraging multidimensional data including laboratory markers (e.g., inflammatory profiles, immune biomarkers), clinical symptoms (e.g., fatigue, dyspnea, neurocognitive changes), and instrumental assessments (e.g., cardiopulmonary function, imaging, gait analysis). The AI-driven system classifies patients by severity and risk trajectories, guiding dynamic care pathways and rehabilitation planning. By aligning LC management with broader NCD prevention strategies and incorporating advanced analytics into Armenia’s e-health infrastructure, this approach enhances system resilience, improves health outcomes, and supports the country’s progress toward the 2030 global health agenda.   read more...   read less

Abstract: Breast cancer is a leading cause of cancer-related mortality for women worldwide. Hormone receptor-positive (HR+) breast tumors, which represent 70% of all breast cancer cases, are treated with endocrine therapy. However, not all patients benefit from this treatment regimen because of patient-to-patient variation. Therefore, high throughput drug screening system is warranted to enable personalized medicine. Patient-derived organoids, which serve as a screening platform, lose their hormone receptors and response upon ex vivo culturing, and may not be adequate models for HR+ breast tumors. Moreover, unidentified biological components in the widely-used basement membrane matrix, Matrigel, result in high batch-to-batch variations and poor reproducibility in organoid cultures. Here, we propose a hydrogel-based 3D ex vivo model with defined structural and chemical properties to test hormone and drug sensitivity of HR+ breast tumors from patient-derived xenografts (PDXs) and patient tumor biopsies using microfluidics. Our data demonstrate the feasibility of this model to preserve cell proliferation and hormone receptor expression over 7 days. We also demonstrate that responses to hormones and FDA-approved drugs are faithfully maintained in this model. Finally, to establish a high throughput hormone and drug testing workflow with transcriptomic readout, we multiplexed barcoded- and drug-treated tumor samples in a single experiment with bulk RNA-sequencing. Our preliminary data demonstrate patient-specific responses to hormones and drugs that correspond to patient genetic mutation profiles, treatment history, disease stages and subtypes. This platform also enables testing of drugs in clinical trials that shows promising therapeutic outcomes for breast cancer, such as CDK4/6i, AKTi, PARPi, mTORi and their combination with endocrine therapy, thanks to the high throughput of the screening system and the low consumption of patient-derived or patient tissue. Given the capability of combining this physiologically-relevant 3D ex vivo model with RNA-seq for HR+ breast tumors, this platform holds potential for high throughput compound testing and transcriptomic profiling of patient biopsies for personalized medicine.   read more...   read less

Abstract: Intravenous iron-carbohydrate complexes are a heterogenous class of nanomedicines whose critical quality attributes (CQAs) have not been fully established. Despite being used in clinical practice for decades, the in vivo biodisposition profiles and mechanisms of biodegradation are not established. Challenges inherent to the bioavailability, distribution and metabolism of this complex class of nanomedicines requires further mechanistic research to support physiologically-based pharmacokinetic (PBPK) models. Bioavailability: Clinical iron indices currently used to evaluate plasma PK profiles of iron-carbohydrate products cannot distinguish between nanoparticle-derived iron in plasma versus endogenous iron. Distribution: Serum iron concentration in plasma does not accurately represent tissue biodistribution. It has been suggested that radio-labeling a commercial iron-carbohydrate preparation may alter their 3- dimensional physical structure, change the nano-bio interface and provide erroneous in vivo distribution data. This greatly limits the understanding the biodistribution in humans. Metabolism: Intravenous iron-carbohydrate complexes do not exhibit a classical drug-enzyme interaction that facilitates the breakdown of the drug into water-soluble species for excretion. Currently, very little is known about the biodegradation mechanism for iron-carbohydrate complexes. However, the rate and extent of the biodegradation process are specific to each unique iron- carbohydrate product. Excretion: Iron is typically highly conserved and recycled and therefore minimal excretion is observed. Excretion is not anticipated to be a key parameter in PBPK models. Conclusion: The fundamental challenge in developing PBPK models is the lack of a fully validated assay to measure nanoparticle-bound iron after administration of an iron-carbohydrate complex. A better understanding of the influences of CQAs on the bioavailability, distribution, and metabolism of the iron- carbohydrate complexes is required. Their complex PK in the context of the highly regulated and dynamic aspects of human iron metabolism necessitates more mechanistic research before PBPK models can be used to describe the in vivo disposition in more detail.   read more...   read less

Abstract: Background: The development and implementation of New Approach Methodologies (NAMs) is crucial for reducing animal testing in drug safety evaluation. However, creating high-quality benchmark datasets with labeled toxicity data remains a critical bottleneck, requiring years of domain expert effort for manual curation. This human-centric approach is unsustainable given the rapid evolution of data sources, with approximately 10-15% of FDA drug labels updated annually, and 40-50 new molecular entities approved each year, demanding years for initial dataset creation and weeks for monthly updates. Methods: We developed AskDrugTox, an automated system leveraging Large Language Models to classify drug toxicity from FDA labeling documents. AskDrugTox processes toxicity endpoints including Drug-Induced Liver Injury (DILI), cardiotoxicity (DICT), and renal injury (DIRI) for over 13,000 marketed drug products, reducing annotation time from minutes per task to seconds. Results: Validation against established benchmarks demonstrated exceptional performance: 93.3% and 94.2% accuracy for DILI and DICT classification after manual verification, with high recall rates over 97.5%. Low inconsistency was reported across same drug products from different manufacturers. The system successfully automated the initial dataset creation in under 100 hours compared to several years manually, while maintaining monthly updates in under 4 hours versus weeks for traditional approaches. Conclusion: AskDrugTox demonstrates the transformative potential of LLMs to generate current, consistent, and accurate drug classification labels through a fully automated and trackable process. This breakthrough enables real-time monitoring of toxicity classification changes due to labeling updates and efficient comparison of labeling documents, significantly accelerating the evaluation and application of NAMs in regulatory science while maintaining human-level accuracy.   read more...   read less

Abstract: Rationale/Background: National drug sales databases are crucial for drug utilization research. Antimicrobials are a regulatory priority due to the risk of multidrug-resistant pathogens from misuse. However, lack of standardization in these databases can delay evidence generation for public health decisions. Objective: To develop a prompt-based approach for creating a dictionary to automatically identify and classify antimicrobials in sales datasets. Methods: An automated classification system was developed using Python and a structured GPT- 4o prompt based on Brazilian antimicrobial prescription regulations and the WHO ATC/DDD system. The model was instructed to analyze each registry entry line-by- line, preserving its structure. The national registry of authorized medicines was pre- processed to remove non-standard products and retain only active, eligible records. GPT-4o classified entries as antimicrobial (yes/no), standardized names, and ATC codes (levels 1–5). This generated a structured dictionary of antimicrobial substances. The dictionary was applied to a 10% sample of January 2020 sales data from Brazil’s SNGPC (Controlled Pharmaceutical Monitoring system). Performance was validated against a manually curated reference list, calculating accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results: Two prompts were tested: Legislation-based only: 90.7% accuracy, 24.4% sensitivity, 99.7% specificity, 90.5% PPV, 90.7% NPV. With explicit antimicrobial list: 100% across all metrics. Conclusion: Large language models can effectively automate complex manual tasks in drug utilization research, such as antimicrobial classification. However, expert oversight is essential to ensure reliability and avoid misclassification. Including a reference list significantly improved recall without sacrificing precision. Well-designed prompts and gold-standard datasets are critical for maximizing performance and reproducibility.   read more...   read less

Abstract: The evolving complexity of medical products along with the integration of advanced technologies demand resilient regulatory science capacity. In response to the progressing needs of this dynamic ecosystem, the Malta Medicines Authority (MMA), as the national competent authority, established its Academy for Patient Centred Excellence and Innovation in Regulatory Sciences - a licensed education institution designing programmes with the aim of strengthening regulatory acumen. Since 2021, the MMA Academy delivered six distinct accredited courses, leading to Awards covering: Good Manufacturing Practice, Good Distribution Practice, Medical Devices, Basic Regulatory Sciences, Pharmacovigilance, and Cannabis for Medicinal and Research Purposes. These educational initiatives converged on European and international legislative and regulatory provisions, blending scientific content with practical application in regulated environments. The multidisciplinary, evidence-based elements embedded in the learning approach intended to address real-world requirements across the diverse contexts, ranging from development to approval, production to supply chain integrity, digitalization, and signal detection to public health considerations. Over 200 local and international participants have engaged with the MMA Academy to date, successfully completing one of the accredited courses offered at different levels of the qualifications framework. Each programme entailed around 25 learning hours, bringing together attendees from an array of settings, mostly professionals with a pharmaceutical background practising in the private/public sector. Feedback forms completed by participants indicated high satisfaction rates across all initiatives. A combined average of 92% of respondents in the evaluative exercise (N=190) stated that they would recommend the courses to colleagues. The strong uptake, interdisciplinary participation and positive outcomes demonstrate the relevance of structured educational offerings and augur a significant impact. The MMA’s strategic investment on the academic front is supporting the cultivation of a forward-thinking community of professionals. These efforts resonate with calls for sustainable commitment and international cooperation in catalysing medical developments through robust regulatory science.   read more...   read less

Abstract: We describe the development and application of a FAIR (Findable, Accessible, Interoperable, Reusable) knowledge infrastructure designed to generate high-quality, transparent, and reproducible evidence to support regulatory submissions incorporating New Approach Methodologies (NAMs). This NAMs Knowledge Infrastructure enables integration of compound data, biological responses, omics datasets, and mechanistic knowledge, structured around pathways and key event relationships. It supports implementation of Next Generation Risk Assessment (NGRA) and Safe and Sustainable by Design (SSbD) principles in practical workflows. Key components include harmonized data templates, compound registration and annotation systems, and knowledge graphs that organize and link biological and toxicological data to support tiered assessment strategies. We implement best practices in omics data management, apply ontological standards, and ensure traceability of results across datasets, models, and workflows. These practices enable robust evidence synthesis and efficient reuse of data across studies, enhancing both scientific validity and regulatory relevance. The infrastructure supports AI-driven applications through structured input data, integration of background knowledge into machine learning and generative AI models, and tools for evidence interpretation. Outputs include regulatory-grade reports with embedded provenance and audit trails, promoting transparency, trust, and reproducibility in assessments. The system is aligned with current and emerging regulatory guidance on the use of digital evidence, AI-based tools, and NAMs, from agencies such as the FDA, US EPA, ECHA, and EFSA. By enabling structured, hypothesis-driven evaluation of chemicals, formulations, and product designs, our approach facilitates the submission of credible scientific evidence that meets regulatory expectations. It strengthens the scientific basis for decision-making while advancing the regulatory acceptance of innovative, non-animal methods. This work has been developed and validated through use cases and case studies from major European initiatives including EU-ToxRisk, RISK-HUNT3R, ASPIS, ACCORDs, SSbD4CheM, and BioPhenom. It is used to support the integrated NAM solutions provided through SaferWorldbyDesign.   read more...   read less

Abstract: EFSA has recently updated its Scientific Guidance on Novel Foods (NFs) to reflect technological and scientific advances. The Guidance highlights how new scientific developments, particularly new approach methodologies (NAMs), could be employed in the toxicological assessment of NFs, with the aim of minimising animal testing in line with the 3R principle. EFSA recommends a tiered approach for absorption, distribution, metabolism and excretion (ADME) and toxicological assessment —covering   genotoxicity and repeated-dose toxicity. The assessment begins with a comprehensive appraisal of existing data on the NF, its source, constituents, and potential metabolites.  A read-across approach based on structurally similar compounds may also be applied where appropriate (e.g., monosodium salt of l-5-methyltetrahydrofolic acid). Depending on the body -of-evidence, in vitro methods using human cell-based systems may also be considered to evaluate absorption, metabolism and genotoxic potential. In case metabolite profiles between human in vitro systems and animals differ considerably, the impact on the running and interpretation of higher tiered in vivo toxicity studies. Since gut microbiota affects the biotransformation of NFs and activation/detoxification of their associated metabolites, in vitro testing using models simulating the dynamic interaction of human gut and its microbiota are included in the first-tier ADME evaluation. Higher-tier in vivo studies (e.g., 90-day study) may be triggered by first-tier findings, following a stepwise strategy. Although NAMs are not yet fully validated for regulatory application, the updated EFSA Guidance on NFs foresees their gradual integration into regulatory toxicology, as validation progresses, to help fill gaps in risk assessment. This approach reinforces science-based decision-making while promoting the 3R principles in animal testing.   read more...   read less

Abstract: Transformer based pre-trained large language models have become ubiquitous. There is increasing evidence to suggest that even with large scale pre-training, these models do not capture complete compositional context and certainly not, the full human analogous context. Besides, by the very nature of the architecture, these models hallucinate, are difficult to maintain, are not easily interpretable and require enormous compute resources for training and inference. Here, we describe Gyan, an explainable language model based on a novel non-transformer architecture, without any of these limitations. Gyan achieves SOTA performance on 3 widely cited data sets and superior performance on two proprietary data sets. The novel architecture decouples the language model from knowledge acquisition and representation. The model draws on rhetorical structure theory, semantic role theory and knowledge-based computational linguistics. Gyan’s meaning representation structure captures the complete compositional context and attempts to mimic humans by expanding the context to a ‘world model’. AI model adoption critically depends on trust and transparency especially in mission critical use cases. Collectively, our results demonstrate that it is possible to create models which are trustable and reliable for mission critical tasks. We believe our work has tremendous potential for guiding the development of transparent and trusted architectures for language models.   read more...   read less

Abstract: Antiviral/antiretroviral therapy (ART) can disrupt intestinal homeostasis, contributing to inflammation as well as metabolic or chronic diseases. Our previous work showed sex-dependent effects of in utero ART exposure on gut homeostasis in adult offspring rats, with male rats showing increased susceptibility to ART-specific changes in the gastrointestinal tract. In this study, we employed a non-animal, human-relevant translational model to investigate the direct impact of ART on male intestinal barrier disruption, inflammation, permeability, and drug transport. Using Microphysiological Systems (MPS)—a New Approach Methodology (NAM)—we developed both organ-on-chip (OC, continuous-flow) and non-continuous (NC, stationary) models from resected human intestinal tissues. Tissues were exposed to ART (low and high doses) using either the OC or NC models for 24 hours, and samples were collected for mRNA gene expression analysis, cytokine/chemokine profiling, and ART transport across the tissue. Transcriptional analysis revealed that baseline gene expression of molecular toxicity related genes (metabolic enzymes: ALDH2, DLD, ECHS1, IDH1, IDH2; DNA repair: APEX1, XPC; transcription factor: HSF1, POU3F3) were lower in the OC as compared to NC. During the ART exposure mitochondrial/detoxification-related gene (ACAD9) and anti-apoptotic gene (BIRC3) were upregulated at the high dose of ART in OC, however, were downregulated in NC, suggesting a possible OC-specific protective mechanism. The proinflammatory cytokines levels (IL-2RF, IL-1A, IL-1B, IL-17, TNF-β, IL-6, MCP-1, MIP-1β) showed no significant changes in OC; but were significantly reduced in male NC tissues following both high and low ART exposures. Notably, model-specific differences in basal/apical drug transport were observed. The OC model demonstrated greater sensitivity to drug/dose variations, suggesting its suitability for capturing physiological transport behaviors more effectively than the NC model. Unlike organoids or 3D cultures, resected gut-on-a-chip model uses minimally manipulated, intact ex vivo human tissues that preserve native architecture, providing a powerful non-animal platform for preclinical risk assessment.   read more...   read less

Abstract: Data modelling methods, such as, machine learning and deep learning models can be used for the classification and prediction of toxicity end points. CNS liability is a major risk in drug discovery and development, leading to delays and failures yet it remains a consideration typically at later stages of the discovery process. Here, we outline the use of different data modelling and data representations methods discussing key considerations when building and selecting such methods for modelling toxicity end points, focused on novel modelling for seizure risk. We have developed a bespoke in-house data set, built on a select group of drug-molecules, chosen based on sub-structure components related to seizure liability, using an electrophysiology assay to determine IC50 values at the chosen ion channels. We present an initial data analysis and preliminary modelling based on this data set, highlighting key early scientific findings this novel modelling approaches.   read more...   read less

Abstract: Accurate, interpretable machine learning (ML) models for predicting clinical outcomes depend on the availability of large, high-quality, structured datasets linking compounds to those outcomes. In practice, clinical trial and post-marketing safety data are fragmented, ambiguously defined, and non-standardized, requiring substantial expert curation. Collapsing diverse clinical manifestations into a single binary label can obscure mechanistic signal and introduce inconsistencies that limit prediction accuracy and interpretability. Drug-induced liver injury (DILI) illustrates this challenge: it encompasses a range of clinical events, from asymptomatic enzyme elevations to acute liver failure, across continuous ranges of severity and frequency. Binary “DILI-positive/negative” labels conflate these presentations, masking relevant dose–response and mechanistic patterns. We developed a granular, dose-dependent, and mechanistically interpretable clinical liver toxicity labeling pipeline based on specific liver-related adverse events (e.g., ALT/AST elevations, bilirubin elevations, hepatitis, acute liver failure). The pipeline integrates large language model–assisted extraction, concrete causal reasoning criteria, standardized statistical thresholds, and rigorous expert validation. Our machine-generated labels achieved ≥95% concordance with human reviewers, and ML models trained to predict specific adverse events achieved similar accuracy to those trained to predict overall DILI risk. They also produced more mechanistically interpretable outputs, particularly when linked to in silico-predicted biochemical, cellular, and ADME features. This approach enables models that are not only more accurate but also more actionable for both biopharma decision-making and regulatory assessment. We propose that adopting such granular labeling frameworks can enhance the transparency, reliability, and translational impact of ML-based clinical risk predictions.   read more...   read less

Abstract: Applying systematic review principles in chemical risk assessment supports a rigorous, transparent and objective analysis of all evidence relating to the assessment task. An important part of the review methodology is the evaluation of internal validity, i.e. the extent to which the design, conduct, and reporting of a study are likely to have prevented bias. A tool for the evaluation of internal validity is needed to incorporate data generated from in vitro studies as evidence in risk assessments. We are applying a systematic and transparent approach to developing such a tool. The tool is called INVITES-IN (IN VITro Experimental Studies INternal validity). Our tool development method includes four sequential studies, involving domain experts and potential end-users of the tool:  Study 1. Creating a comprehensive database of items (“item bank”) that may affect the internal validity of in vitro studies of any design from currently available risk of bias tools and three focus group discussions with in vitro experts.   Study 2. Prioritising items from the item bank according to their importance to the assessment of internal validity of in vitro studies using a modified Delphi approach.   Study 3. Creating a beta version of an appraisal tool based on analysis of the prioritised items identified in stage 2.    Study 4. Creating a final release version of the tool, based on user-testing of the beta version.  INVITES-IN will provide a structured approach for retrospective internal study validity assessment. It will i) support objectivity, transparency and openness in the process of summarizing evidence for decision making, ii) support the incorporation of evidence from in vitro data in chemical risk assessments and pave the way for the use of such data as stand-alone evidence (both guideline and non-guideline studies), iii) contribute to confidence in the use of NAMs data as evidence in chemical risk assessments.   read more...   read less

Abstract: The development of the medical software immuneS addresses the growing need for intelligent digital tools that support clinical decisions and provide health information to citizens. immuneS serves as a decision support system for diagnostics to clinicians and an informational interface for citizens. This dual functionality presents complex challenges in software engineering, quality management, and regulatory compliance within the framework of in vitro diagnostic medical devices in Switzerland and Europe. A key challenge in the development of immuneS lies in integrating artificial intelligence (AI) components into a robust Quality Management System that aligns with ISO 13485, CH IvDO 812.219, and EU Regulation 2017/746 (IVDR). AI algorithms are essential for diagnostics accuracy and clinical workflow efficiency with large-scale data. However, they introduce issues around transparency, reproducibility that conflict with the static validation paradigms traditionally required for medical software. Ensuring that these AI-driven features remain interpretable, clinically reliable, and traceable throughout the product lifecycle is crucial, particularly when assessing clinical performance in line with IVDR requirements. Furthermore, medical device software must comply not only with technical and safety standards but also with strict documentation and performance evaluation protocols. The classification of immuneS as a Class B mandates rigorous demonstration of clinical benefit, analytical performance, and scientific validity. Achieving compliance requires a tightly integrated development and regulatory strategy. The technical, regulatory, and practical considerations involved in developing immuneS are outlined and the intersection of AI with medical software compliance frameworks is highlighted. This minimal-example experience provides insights into the future of intelligent diagnostic systems and their safe integration into European healthcare.   read more...   read less

Abstract: Purpose This research aims to explore the implications of Paragraph IV certification on generic drug market entry while innovators' patents are still active. It also examines the challenges posed by patent settlements, the BLOCKING Act, and the FDA's decision-making process regarding generic applications. The objective is to shed light on the complexities of pharmaceutical market dynamics, regulatory practices, and the balance between innovation and accessibility. Methods The study involves a detailed analysis of the USFDA's recently published Para IV certification list and application statuses from 2020 to 2024. It reviews litigation trends among top generic manufacturers and scrutinizes the impact of patent disputes on market entry. The study also evaluates the potential effects of the BLOCKING Act on the 180-day exclusivity period and generic drug market dynamics. Results The analysis reveals that nearly half of the ANDA applications submitted through Paragraph IV certification are deemed 'Eligible,' with significant portions either deferred or extinguished. A peak in eligible ANDA submissions was observed in 2023. Litigation trends show active patent challenges by leading generic companies like Teva, Apotex, and Actavis. The study highlights the FDA's role in facilitating first ANDA approvals, emphasizing the importance of providing safe and effective generic alternatives. Conclusion The study concludes that while the FDA supports generic companies in their application processes, patent litigation remains a significant hurdle, delaying the availability of cost-effective generic drugs. The BLOCKING Act, if enacted, is likely to disrupt the 180-day exclusivity incentive, reducing the predictability and value of generic drug market entry.   read more...   read less

Abstract: In 2018, the EPA announced that it would discontinue funding research using animals by 2035. In December 2022, the FDA Modernization Act 2.0 was signed, which explicitly states the facilitation of NAMs' development as alternative animal testing methods, including MPS. MPS (Microphysiological System), which reconstructs an in vivo-like culture environment in a microscopic space, is being considered for applications in toxicity, safety and efficacy testing of various modalities of pharmaceuticals. The MPS is expected to be a useful tool for the development of new drugs, as it can mimic the functions of various organs, such as the liver and small intestine, and can use human cells to evaluate efficacy and safety on humans. For MPS, in addition to the cell/tissue culture technology and the device technology for holding and measuring cell/tissues, standardization is essential so that users can freely construct MPS in accordance with their CoU and conduct tests under precision control. International standardization relating to devices and cells, tissues, etc., has been underway. International standardization of devices has been carried out in a committee, ISO/TC 48, which has been working on microfluidic devices. For example, ISO 10991 defines terms related to microfluidics, ISO 22916 describes device dimensions and connections, and other standards have been published. Meanwhile, standardization of biological components such as cells and tissues has been discussed in CEN/CENELEC Focus Groups, and since the publication of the Load Map last year, a structure change of ISO/TC 276 (Biotechnology) has been performed and move the discussion of biological components into ISO. In this report, the status of international standardization of MPS, which is expected to become one of these NAMS, will be presented.   read more...   read less

Abstract: The present poster will present a PhD research that is studying the phenomenon of the non- adoption of innovation in the context of health regulatory agencies that are implementing Artificial Intelligence (AI) projects in their internal processes. Although there is much research on the diffusion of innovation, especially focusing on the implementation phase, organizations have still encountered difficulties during this process. These difficulties have been studied mainly by innovation resistance frameworks. However, there is still a gap in this field of knowledge that affects the day-to-day activities of organizations. Thus, the present research aims to study the adoption of an innovation in the internal processes of an organization that has decided to adopt an innovation but finds non- adopters. In this sense, the research seeks to answer the question: why are there non-adopters within an organization that has already decided to adopt an innovation? Starting from a multiple- case study design, the study conducts qualitative research in two agencies, Swissmedic (Switzerland) and Anvisa (Brazil), that are running AI projects focused on their internal processes. Data were collected through interviews. This research corroborates and extends previous research regarding the literature on diffusion of innovation with the study of non-adoption, looking to understand the reasons behind non-adoption and the factors that contribute to it. The novelty lies in AI project analysis in the internal processes of public organizations. Besides that, the research seeks to generate knowledge that could be useful for these organizations in implementing their projects. Data analysis shows that organizations still do not know exactly which problems AI is the most effective solution for, but they feel pressured to adopt it, an illustration of the garbage can model. In addition, power conflict and public values also stood out. The data analysis indicates that perception and personal goals are possible explanations for non-adoption.   read more...   read less

Abstract: The use of artificial intelligence (AI), and in particular large language models (LLMs) such as those behind tools like ChatGPT and Google Gemini, is expanding within toxicology applications. However, in areas related to toxicological risk assessment (where conclusions have direct implications for public health), such advancements must progress with attention to accuracy, explainability and accountability. Explainability is generally considered a critical factor for building trust in health-based decision-making, with transparency serving as a critical means to achieve it. Accountability, in turn, is essential for regulatory acceptance and typically arises from transparent processes and defensible expert review. This study examines the potential of LLMs to support the derivation of Permitted Daily Exposure (PDE) values (a compound-specific daily dose that is unlikely to cause an adverse effect for a lifetime daily exposure). A case study compares the results from AI- assisted processes and manual approaches for PDE derivation. The proposed AI-assisted approach integrates LLMs into a structured workflow covering: data discovery, knowledge extraction, integrated analysis and summarization, data interpretation and review. LLM’s speed and scalability improve PDE derivation, but the workflow enables expert review to mitigate limitations of LLMs and the black-box behavior (e.g., hallucinations). This work encourages the creation of AI-assisted frameworks for the responsible application of LLMs in toxicology, where expert review is not simply a safeguard but a critical component. It guarantees that AI augments, rather than replaces, expert judgment in regulatory decision making. This human-centric AI-assisted approach keeps accountability for risk assessment with the experts and leverage AI’s capabilities. The results is a pragmatic and responsible use of LLMs for toxicological risk assessment, allowing humans (e.g., scientists and regulators) and AI to collaborate, each contributing to the outcome with their strengths and capabilities.   read more...   read less