An AI Foundation Toxicology Model and Framework to Support Waiving a Second Species in Drug Safety Studies

Expected Impact

The action under this topic is expected to achieve the following impacts:

• Faster and more informed decision-making through the use of an AI-driven NAM (AI Foundation Toxicology Model) and increased efficiency through rapid processing of vast amounts of data [1].
• Increased consistency and standardisation in a NAM-based approach, specifically an AI model, used by industry in the efficient development, testing and production of safe and effective innovative health technologies, improving industrial competitiveness.
• Regulatory adoption of a NAM-enabled second species waiver model (AI Foundation Toxicology Model) and weight-of-evidence framework, in line with recommendations and more consistent global decision-making on waiving second species testing.
• Reduction in animal use, accelerated timelines and lower costs, enhancing the competitiveness of the European health industry through economical and ethical benefits.
• Improved public health as patients will benefit from safe and effective medicines developed faster using validated NAMs.

The action is expected to contribute to the EU Directive (2010/63/EU) [2] on the protection of animals used for scientific purposes and the implementation of the 3Rs principles to replace, reduce and refine the use of animals. The action is expected also to consider and contribute to EU programmes, initiatives and policies on New Approach Methodologies (NAMs) such as the future European Research Area (ERA) action on accelerating NAMs to advance biomedical research and testing of medicinal products and medical devices.

[1] Transforming animal study toxicology reports into structured, harmonized data using large language models

[2] European Directive on the protection of animals used for scientific purposes

Expected Outcome

The action under this topic must contribute to all of the following outcomes:

• A validated Artificial Intelligence (AI) Foundation Toxicology Model* that provides transparent probabilistic predictions for industry and regulator stakeholders to determine when a second species in chronic (>90 days) and sub-chronic (90 days) small molecule medicine repeat-dose studies is unlikely to provide additional safety relevant information, including risks of missed toxicity, organ-specific findings, and divergence in No Observed Adverse Effect Level (NOAEL). The goal would be to enable waiving the need for two species chronic testing for small molecules and other modalities e.g. oligonucleotides.
• A standardised, transparent weight-of-evidence framework for industry, regulator and academic stakeholders that enables reproducible assessment of evidence quality, consistency, relevance, and uncertainty across regulatory submissions, supporting the wider adoption of the AI Foundation Toxicology Model and New Approach Methodology (NAM)-based toxicology strategies in general.
• Functional tools, templates, and training materials that support the real-world implementation, sustainability and evolution of the foundation model and weight-of-evidence framework including guidance on explainability, provenance, governance, ethical use, and alignment with AI requirements, tailored to industry, regulator and academic stakeholder needs.
• Enhanced industry and regulator stakeholder confidence in second species waiver applications, particularly for small molecule medicines, supported by empirical, calibrated evidence and a framework enabling predictable adjudication, more consistent global waiver decision-making and timely progression of medicine development without compromising patient safety. This confidence should be gained through the model and framework’s application for regulator validation and acceptance, with the longer-term goal, beyond the action’s scope, of a revision of the regulatory guidelines ICH M3(R2) [1] taking onboard the future project’s outcomes. This topic should provide the opportunity to extend this confidence to waiving chronic testing in a single species beyond small molecule medicines and to other study types.

* An AI Foundation Toxicology Model is a large-scale, pre-trained machine learning system that learns generalisable representations of toxicological biology and chemistry from diverse, multimodal datasets, enabling broad applicability across safety assessment tasks with minimal task-specific training.

[1] ICH M3(R2). Nonclinical safety studies for the conduct of human clinical trials and marketing authorization for pharmaceuticals. In: International Conference on Harmonisation (ICH). Topic M3(R2); 2009.

Scope

Regulatory guidelines e.g. ICH M3(R2) for nonclinical safety assessment of new small molecule medicines have historically required toxicity studies in two species, typically one rodent and one non-rodent, to maximise the likelihood of detecting adverse effects, improve the translatability to humans and guide patient dosing and monitoring. This two species approach is increasingly challenged by ethical considerations, long and costly study durations, limited reproducibility, and uncertainty regarding human relevance [1], [2]. In parallel, scientific progress in mechanistic toxicology, advanced in vitro systems (e.g. organoids and micro-physiological systems), multi-omics technologies, and AI based modelling is enabling NAMs to help reduce reliance on animal models without compromising patient safety. Within the context of this topic, the AI Foundation Toxicology Model will be considered a NAM, facilitating NAM-enabled second species waivers.

In addition, the policy landscape across global jurisdictions is evolving rapidly. The European Commission is preparing a roadmap towards phasing out animal testing for chemical assessment safety, expected in 2026 [3]; the European Medicines Agency (EMA) has increased its focus on emerging NAMs through horizon-scanning and dedicated working groups [4]; the UK Government has published the Replacing Animals in Science strategy [5] and the United States Food and Drug Administration (FDA) has published a dedicated roadmap for reducing animal testing in preclinical safety studies [6]. Collectively, these initiatives highlight the need for robust evidence, data integration, validated methods and coordinated stakeholder action to support credible alternatives to animal testing.

A weight-of-evidence approach can eliminate unnecessary animal tests by integrating and assessing diverse data sets as well as determining whether the available evidence is sufficient or additional nonclinical testing is required to address potential safety concerns. The value of this evidence depends on its quality, consistency, human relevance, and the nature and severity of observed effects. Currently, the absence of standardised, data-driven methods means decisions can vary across reviewers and jurisdictions, creating an uncertainty which discourages changes from being made to the current process. A more structured, reproducible, validated, and evidence-based approach would support consistent, transparent, and well-justified decision-making on whether second species studies could be waived.

This topic aims to address and standardise the waiver process for second species chronic and sub-chronic small molecule general toxicology studies by developing a transparent, AI-driven Foundation Toxicology Model. The model should be underpinned by diverse, standardised, high quality in vivo datasets, encompassing all modalities of medicines, and supported by a robust weight-of-evidence framework. The most appropriate single species should be selected early in the drug development programme based on target affinity and functions, metabolite profiling, etc. The main context of use of the AI Foundation Toxicology Model is to predict whether testing in a second species will provide additional safety-relevant information. If toxicity predictions are equal across rodent and non-rodent species, priority should be given to the reduction of non-rodent chronic toxicity testing.

Although the context of use for this topic should be focused on waiving second species testing for chronic and sub-chronic small molecule studies, the project’s actions, sustainability plan plus the ongoing development and evolution of the model should provide the opportunity to assess appropriate single species selection and encompass the broadening of the context of use to enable the waiving of single species chronic testing in small molecules and other modalities, such as oligonucleotides. Therefore, relevant data sources and their provision should enable this broader and longer-term goal. Selection of the appropriate single animal species for preclinical safety evaluations for biotechnology-derived pharmaceuticals in line with ICH S6 [7] should be considered within the scope of this project if the relevant data provision is sufficient to enable this functionality.

This topic should build upon, among others, the outcomes, learnings, data sharing structures, and methodologies of previous IMI/IHI projects that improve the predictability, feasibility and reliability of pre clinical safety assessments — such as BigPicture, eTRANSAFE , eTOX , imSAVAR , and VICT3R — as well as other relevant European, international and national initiatives such as Animal free safety assessment of chemicals: Project cluster for implementation of novel strategies (ASPIS) and NC3Rs-led efforts like the Virtual Second Species [8] initiative and the Two Species project [9]. Preliminary analysis from the NC3Rs Two Species project predicts that it would be possible to avoid conducting studies in non-rodents (e.g. dog or non-human primate) in >60% of the small molecule drugs that enter Phase II globally each year [10].

By fostering collaboration among pharmaceutical companies, regulatory agencies, academic institutions, SMEs, and patient advocacy groups, the action generated by this topic should seek to enhance patient safety, reduce reliance on animal testing, streamline medicine development timelines, lower associated costs, and boost the global competitiveness of the pharmaceutical industry within the European Union.

To fulfil this aim, the proposal should:

1. Consolidate and evaluate data sources

• Identify, curate, and assess the quality and suitability of data from multiple pharmaceutical companies and other organisations within the consortium with a priority focus on general toxicology studies (14-day, 28-day, 90-day and chronic toxicity studies) in the form of structured toxicity data in CDISC SEND [11] format as well as in unstructured study reports. Secondary focuses for additional data should also be considered, including in vivo toxicity study data, chemistry (e.g. structure information, where permitted), pharmacokinetic and exposure data, mechanistic in vitro systems, in silico models, multi-omics datasets, licensed content, prior consortia outputs, and public regulatory databases.
• Develop a new database to host and analyse data, encompassing principles to support a common or federated data model that enables sensitive data preservation and multisource analysis that can allow NAM-enabled second species waivers. Additionally, sensitive data should undergo various levels of blinding from full blinding to structural alerts as a surrogate for the full structure and structural embeddings with noise introduction among other approaches.
• Applicants are expected to consider the potential regulatory impact of the project’s results and develop a regulatory strategy and interaction plan for generating the required evidence to support regulatory decision-making, linked to the data sources used, as well as engaging with regulators in a timely manner for their input (e.g. national competent authorities, EMA, etc). The goal is to gain, by the end of the project, endorsement of the AI Foundation Toxicology Model from the regulatory authorities (e.g. EMA, FDA, including potential Innovative Science and Technology Approaches for New Drugs (ISTAND) [12] submission) to enable second species waivers for chronic and sub chronic small molecule testing. The longer-term goal, beyond the action’s scope, of revising the regulatory guidance ICH M3(R2) in line with this project’s outcomes should also be taken into consideration when designing the regulatory strategy. The opportunity to broaden the scope of the AI Foundation Toxicology Model to encompass waiving single-species chronic testing beyond small molecules (e.g. oligonucleotides) should be part of the longer-term and sustainability planning.

2. Select and optimise AI methods

• Conduct a systematic review of existing AI and AI-supported foundation modelling approaches.
• Select and optimise AI methods able to perform probabilistic predictions estimating the likelihood of novel safety-relevant information being identified in longer-term studies and/or in a second species, including risks of missed toxicity, organ-specific findings, and divergence in NOAEL. The potential to select the appropriate single species for biological medicines should also be considered in the AI method selection and optimisation.
• Ensure transparency, interpretability, traceability, and data provenance to meet regulatory expectations.

Applicants should ensure that, in order to support regulatory acceptance and the scientific robustness of the AI Foundation Toxicology Model, it is built using a structured and transparent AI methodology encompassing:

• Data harmonisation and representation learning
• Feature extraction pipelines appropriate to each data modality (unstructured study reports, molecular structures, omics datasets, mechanistic assays).
• Multimodal representation learning (e.g., contrastive learning) to create unified latent spaces across biological, toxicological, and metadata features.
• Federated learning and privacy preserving techniques to enable secure, multi organisation data contributions without exposing proprietary information.
• Predictive modelling approaches
• Probabilistic supervised models (e.g., Bayesian neural networks, Gaussian processes, calibrated ensembles) to estimate the probability of missed toxicological findings under a single-species approach.
• Mechanistic-informed modules integrated via knowledge graph neural networks, causal inference frameworks, and multi-task predictors for endpoints such as drug induced liver injury, cardiotoxicity, and genotoxicity.
• Uncertainty quantification methods to ensure transparent confidence intervals suitable for risk-based decision-making.
• Scenario simulation engines (Monte Carlo, generative models, causal models) to test counterfactuals relevant to second species value.
• Transparency and explainability- Use of explainable AI techniques to identify key drivers of model predictions.- Comprehensive provenance tracking to ensure every model output is traceable.- Human interpretable decision layers to support weight-of-evidence narratives.
• Long term evolution
• Ability for continuous learning, single-species chronic testing waivers beyond small molecules and modular updates as new data types and methods emerge.

In parallel, other approaches should be assessed in case a foundation model does not provide the solution. This could encompass decision-tree approaches, Bayesian statistical methods and/or more classical Machine Learning (ML) approaches.

3. Develop and validate the AI Foundation Toxicology Model

• Design, train, and validate an AI Foundation Toxicology Model that produces probabilistic predictions across multiple dimensions: 1) study duration (sub-acute, 14–28 days; sub-chronic; and chronic), 2) species (rodent and non-rodent), and 3) outcome type including the likelihood of identifying a novel target organ, a significantly divergent toxicity profile and/or a significantly divergent NOAEL. For example, the model should predict the likelihood (0-100) that running a 9-month non-rodent study would identify a divergent toxicity profile, including significantly increased severity of pathology findings as compared to the toxicity profile identified in earlier, shorter-term studies or projected to be identified in the rodent 6-month study yet to be conducted. These specific outputs should be defined depending on the data provided and the AI methods, which will be established within the early stages of the project.
• Industry, academic and SME stakeholders with consistent regulator engagement and input should define representative use cases and performance metrics, including accuracy, robustness, explainability, extensibility and trustworthiness, to validate the model. Formal benchmarking and rigorous testing are essential for ensuring consistency of results, to build trust in the model's recommendations.

4. Establish a standardised weight-of-evidence framework

• Create a transparent, reproducible, weight-of-evidence decision support framework that integrates diverse data types and model outputs into standardised reasoning steps, including the direct model outputs (e.g., probability and significance of novel findings in a longer-term second species study) and model context (e.g., narrative rationale supporting the probabilistic output) from the AI Foundation Toxicology Model as well as orthogonal internal programme and external literature support. In vitro NAMs, target knowledge and literature mining would all help supplement and increase confidence in the AI Foundation Toxicology Model outputs and recommendations.
• Define how uncertainty, evidence quality, consistency, severity, and human relevance including potential impact on or risk for the clinical programme should be evaluated, compatible with regulatory reasoning, to support second species waiver submissions.
• Test the AI Foundation Toxicology Model, through use cases, together with the weight of evidence framework with relevant diverse stakeholder groups, including pharmaceutical and biotechnology companies, regulators and academics/SMEs working at the interface of drug safety research and regulation to ensure consistency across reviewers and jurisdictions.

5. Prepare for regulatory uptake and long-term sustainability

• Develop recommendations and practical tools for real-world implementation, including regulatory strategy and guideline revision as required, alignment with ethical and legal principles plus a governance structure for ongoing model evolution and ecosystem adoption. For instance, trustworthy AI, human oversight and verifications will follow regulatory frameworks such as the Assessment List for Trustworthy Artificial Intelligence (ALTAI) [13] .
• Devise a sustainability and evolution plan for long-term hosting, maintenance, continuous data harmonisation and improvement of the AI Foundation Toxicology model and weight-of-evidence framework, enabling broad and sustainable ecosystem adoption.

[1] Prior et al. (2020). Opportunities for use of one species for longer-term toxicology testing during drug development: a cross industry evaluation. Regulatory Toxicology and Pharmacology. 29;113:104624

[2] Prior et al., (2024). Exploring Greater Flexibility for Chronic Toxicity Study Designs to Support Human Safety Assessment While Balancing 3Rs Considerations International Journal of Toxicology. 43(5):456-463.

[3] European Commission – Roadmap towards phasing out animal testing

[4] EMA – New Approach Methodologies Horizon Scanning Report

[5] UK Government – Replacing Animals in Science Strategy

[6] FDA – Roadmap to Reducing Animal Testing in Preclinical Safety Studies

[7] Preclinical Safety Evaluation of Biotechnology-Derived Pharmaceuticals

[8] NC3Rs Virtual Second Species - Applying advanced computational and mathematical modelling approaches to develop a suite of virtual dog tissues and organs to model toxicological endpoints for new chemical entities

[9] Review of the use of two species in regulatory toxicology studies | NC3Rs

[10] Passini et al (2025). OS02-10 Analysis of the use of two species in regulatory toxicology studies for molecules following ICH M3(R2) Toxicology Letters (EUROTOX 2025)

[11] SEND specifies a way to collect and present non-clinical data in a consistent format, a requirement for data submission to regulators

[12] Innovative Science and Technology Approaches for New Drugs (ISTAND) Program, FDA

[13] Assessment List for Trustworthy Artificial Intelligence (ALTAI)