The field of exposomics, dedicated to understanding how environmental exposures impact human health, requires extensive data harmonization across numerous disciplines and vast timelines, tracing entire life courses rather than isolated events or exposures. Until now, integrating heterogeneous datasets and disparate research findings has been a persistent bottleneck. The emergence of ToxIndex offers an agentic system capable of autonomously navigating and consolidating these vast sources of toxicology data, greatly enhancing the capacity for integrative and comprehensive risk assessment. This paradigm shift holds promise to elevate the rigor and speed of scientific discovery in exposomics and beyond.

Historically, safety testing for chemicals has been exorbitantly expensive, consuming nearly $20 billion annually, yet only a fraction of chemicals in commerce—approximately 10%—have undergone any form of safety evaluation, and a mere 1% receives thorough scrutiny. The 2007 landmark National Research Council report, Toxicity Testing in the 21st Century, envisioned supplanting traditional animal-testing approaches with computational models, in vitro assays, and mechanistic pathway analyses. Over the past two decades, the scientific community has developed extensive New Approach Methodologies (NAMs), ranging from validated quantitative structure-activity relationship (QSAR) models to high-throughput screening initiatives like ToxCast and Tox21, alongside curated adverse outcome pathways and regulatory databases.

Despite these advances, a critical integration layer that unifies the vast output and capabilities of multiple NAMs has been elusive—until now. ToxIndex, conceived and developed by Dr. Thomas Luechtefeld under Dr. Hartung’s mentorship at CAAT, addresses this deficiency directly. The platform leverages AI agents to programmatically access, synthesize, and orchestrate a comprehensive suite of toxicological data sources, enabling unified and audit-ready toxicological workflow generation. This integration spans computational predictions, in vitro bioassays, and physicochemical as well as absorption, distribution, metabolism, and excretion (ADME) profiles to deliver fully contextualized risk evaluations.

At its core, ToxIndex incorporates three interlinked tiers of evidence. First, the in silico component draws on over 600 containerized open-source toxicology models, enhanced by a proprietary transformer architecture trained on an unprecedented dataset of 254 million human chemical activity measurements. Second, in vitro evidence is extracted from a massive knowledge graph containing upwards of 60 billion data triples, co-developed with the National Toxicology Program’s NICEATM under NSF funding. This graph integrates renowned sources such as ToxCast, Tox21, ChEMBL, and PubChem, alongside 1,200 meticulously curated datasets. Third, physicochemical and ADME properties are supplemented through expert-curated sources including REACH dossiers, EPA databases, and tools like the OECD QSAR Toolbox, rounding out the platform’s comprehensive evidence base.

A hallmark of ToxIndex is its granular provenance tracking. Each risk claim generated by the platform is fully auditable, referencing specific source datasets down to the exact database table, row, and column. This meticulous traceability not only satisfies but frequently surpasses existing regulatory requirements such as the stringent documentation standards of REACH and TSCA. By ensuring transparency and accountability, ToxIndex meets the critical need for reproducibility and regulatory compliance in toxicological assessments.

In a rigorous proof-of-concept demonstration, ToxIndex successfully produced a 47-page toxicological risk assessment for dodecanedioic acid (DDDA), encompassing 944 individual claims, within a mere three hours. This evaluation, which would traditionally demand several months of labor-intensive expert review, now emerges instantly from the platform. What is more, each claim is scored using the Klimisch system for reliability, facilitating evidence weighting and regulatory acceptability. The platform’s dynamic document crawler also continuously updates evaluations as novel scientific literature and regulatory texts are indexed globally, ensuring assessments remain current without manual intervention.

ToxIndex features adaptive capabilities that allow it to reassess chemicals dynamically as new data become available. It can prioritize evidence gaps and automatically route requests for additional experimental testing to laboratories. Furthermore, it maintains an evolving knowledge graph where every new NAM and data source is registered and integrated, creating synergistic value as each contribution enhances the utility of others. This creates a living, interconnected toxicology landscape that fosters robust, up-to-date safety profiles.

The innovation arises from a transatlantic research partnership. Dr. Hartung and Dr. Luechtefeld also collaborate within ONTOX, an ambitious €17.2 million European Horizon 2020 project focused on non-animal toxicity prediction strategies. Through ToxTrack LLC—Insilica’s sister company—ToxIndex has already been adopted by regulatory scientists in the European Union and rigorously validated against international standards. Public release plans include sharing scientific methods via arXiv, distributing code on GitHub, and broadly disseminating the underlying transformer AI technology. Engagements with NIH and OECD are underway to embed ToxIndex into global regulatory frameworks, promising its recognition as essential toxicological infrastructure.

The timing of ToxIndex’s public debut could not be more critical. Regulatory landscapes worldwide are rapidly evolving to embrace non-animal, AI-driven methodologies. Landmark policy changes such as the FDA Modernization Act 2.0 have formally eliminated animal testing mandates and explicitly endorse computational alternatives. Simultaneously, the EPA aims to eradicate mammalian testing by 2035, and Europe grapples with the massive compliance load imposed by REACH and bans on animal testing in cosmetics. This convergence of science, technology, and policy demands scalable, transparent, and high-throughput toxicological assessment tools like ToxIndex.

The surge in AI-driven drug discovery compounds this need, as thousands of novel molecular entities synthesized algorithmically require thorough safety evaluations. With only around 9,000 toxicologists serving North America and regulatory review timelines already stretched beyond legal limits for 88% of new chemicals, traditional processes are untenable. Insilica’s platform offers a practical solution to scale toxicity evaluation, avert bottlenecks, and accelerate the development of safe and effective chemical products.

Dr. Hartung’s extensive regulatory experience, including seven years as head of ECVAM at the European Commission and advisory roles with the EPA, FDA, EFSA, OECD, and Apple’s Green Chemistry Advisory Board, provides the essential credibility and trust foundation for Insilica’s collaboration with agencies and industry. He emphasizes the indispensable need for transparency and validation asserting, “95% accuracy is not good enough when human lives are at stake. Trust must be earned through institutional relationships, transparent methodology, and rigorous validation.” This ethos underscores the platform’s integration of scientific rigor, traceability, and transparency.

Both Dr. Hartung and Dr. Luechtefeld have expressed enthusiasm about ToxIndex’s potential. Dr. Hartung remarks that while the scientific groundwork—including databases, models, and policy advocacy—has long existed, ToxIndex embodies the engineering breakthrough required to integrate these components at regulatory scale, marking a transformative moment in toxicology. Dr. Luechtefeld, a CAAT alumnus, credits his mentor and states that the essence of 21st-century toxicology lies not in building isolated models but in building a system that harmonizes every resource, establishing ToxIndex as this vital systemic solution.

Insilica plans to showcase ToxIndex at the upcoming Society of Toxicology (SOT) 65th Annual Meeting in San Diego, slated for March 22–25, 2026. Attendees will experience live demonstrations featuring real-time AI-driven risk assessment, automated IUCLID dossier generation for REACH compliance, and immersive interaction with the Brickyard knowledge graph. Both Dr. Hartung and Dr. Luechtefeld will be available for in-depth presentations and media engagements during the conference, inviting close scrutiny and adoption by the toxicology community.

Founded in 2017 by Dr. Luechtefeld, Insilica embodies a full-stack AI toxicology firm, blending cutting-edge machine learning, expert toxicological insight, and the world’s most extensive toxicology data infrastructure. Serving a diverse client base spanning pharmaceutical, energy, agriculture, cosmetics, and consumer product industries, Insilica operates as an official contractor to the FDA and partner in the European ONTOX consortium. Based in Rockville, Maryland, the company stands at the forefront of ushering toxicology into an era of integrative AI-driven safety evaluation.

The Johns Hopkins Center for Alternatives to Animal Testing (CAAT), founded in 1981 and directed by Dr. Hartung, remains a trailblazer in advancing humane science by developing and validating non-animal methods. CAAT fosters computational toxicology innovation and serves as the U.S. hub for the Transatlantic Think Tank for Toxicology (t4), collaborating closely with CAAT-Europe. Their pioneering work in evidence-based toxicology and computational approaches has laid the foundation for platforms like ToxIndex, transforming the future landscape of chemical safety and regulatory science.

Subject of Research: AI-driven toxicological risk assessment and integration of New Approach Methodologies (NAMs) in chemical safety evaluation

Article Title: Transforming Toxicology: Launch of ToxIndex, an AI Platform Delivering Regulatory-Grade Chemical Risk Assessment in Hours

News Publication Date: March 14, 2026

Web References:
https://insilica.co
https://sage.toxindex.com
https://caat.jhsph.edu

Image Credits:
3rd Edition of the International Conference on Advances in 3D Cell Culture held in conjunction with the 6th Annual Summit on Biopharmaceutical Product Development | Jan 22-23, 2026 | Goa, India

Keywords: Artificial intelligence, Toxicology, Environmental toxicology, Omics, Metabolomics, Computational science, Drug development, Big data, Bioactive compounds

Artificial intelligence is transforming various sectors, and its implications for public health are particularly promising. The study delineates how AI applications, including conversational agents and generative technologies, can break down barriers between researchers and community members. Through mechanisms such as large language models, AI systems can facilitate dialogue that is more engaging and accessible for diverse audiences. This increased engagement is crucial, as it ensures that the voices of underrepresented populations are heard and included in the scientific process.

Conversational AI stands at the forefront of these innovations, fostering an environment where community scientists can express their concerns, share insights, and collaborate with researchers in a meaningful way. The use of conversational AI can democratize the research process, allowing for more inclusive contributions to the development of health interventions tailored to local needs. This is pivotal in addressing health disparities prevalent in marginalized communities, where access to information and resources is often limited.

Alongside conversational AI, the study emphasizes the significance of generative AI in public health research. Techniques such as text-to-image AI can revolutionize how data is presented and perceived by the public. Visual aids can make complex research findings more digestible, thereby improving public understanding and facilitating more informed discussions about health issues. The role of generative AI in data visualization cannot be overstated, as it bridges the gap between intricate scientific outcomes and community comprehension.

Moreover, the promise of predictive analytics powered by AI further enhances the capability of citizens to engage in scientific research. By analyzing extensive datasets, AI tools can identify emerging health trends and potential risks, encouraging proactive measures within communities. This foresight is invaluable, enabling public health officials and community members alike to anticipate challenges that may impact their health and well-being and respond effectively. By empowering individuals with the knowledge gleaned from predictive analytics, communities can take ownership of their public health narratives, shaping the future in a more informed manner.

However, the study does not shy away from addressing the ethical complexities intertwined with the incorporation of AI in citizen science. As AI systems are developed and deployed, the importance of mitigating biases that may arise in algorithmic decision-making becomes paramount. Researchers emphasize that the tools intended to empower communities must not inadvertently reinforce existing inequalities or introduce new biases. The ethical implications of AI technologies necessitate a thoughtful approach to their design, implementation, and evaluation, ensuring that they serve equity rather than undermine it.

Data privacy is another critical concern raised by the researchers. As AI systems analyze vast amounts of information, safeguarding the privacy of individuals participating in research must be a foundational principle. Building trust with communities is essential for successful and ethical implementation of AI in citizen science. Researchers underscore the need for transparency in how data is collected, processed, and utilized, promoting informed consent and educating community members about their rights.

Ongoing community engagement is also vital in developing and maintaining ethical AI frameworks in public health research. The study advocates for continuous dialogues between researchers and community leaders to identify potential ethical dilemmas and collaboratively develop solutions. Engaging with communities not only fosters trust but also enriches the research process by incorporating local knowledge and expertise, thereby enhancing the relevance and impact of health initiatives.

In light of the potential benefits and risks associated with AI in citizen science, the authors of the study have provided additional resources to elucidate key points in their research. A video discussion accompanies the publication, allowing researchers to communicate their findings in an accessible format, thereby broadening the reach of their work. This multimedia approach exemplifies the innovative spirit of contemporary research methodologies that leverage technology to enhance understanding and engagement.

As the dialogue around AI in public health evolves, the researchers are contributing to a broader conversation about the role of technology in societal advancement. With a commitment to ethical research practices and respect for community voices, the study aims to inspire future explorations at the intersection of AI, public health, and citizen science. The potential of AI to drive meaningful change is significant, yet the responsibility of researchers remains to ensure that this technology is harnessed for the collective good.

The implications of this study extend beyond mere academic interest; they raise critical considerations for policymakers, public health officials, and community leaders as they navigate the rapidly changing technological landscape. By prioritizing health equity and actively engaging communities in the research process, stakeholders can facilitate a healthier future for all, empowered by the collective wisdom of both technology and society.

This research not only highlights the boundaries of current AI applications but also inspires further inquiries into the uncharted territories that lie ahead. The future of health equity in our communities hinges on embracing innovations while remaining vigilant in their ethical deployment, ultimately assuring that advancements in technology translate into tangible benefits for all.

As this study illustrates, the confluence of AI and citizen science presents unique opportunities that can redefine public health paradigms. Through responsible stewardship of these innovations, researchers can help guide the evolution of citizen science towards greater empowerment, inclusivity, and, ultimately, a more equitable health landscape.

Subject of Research: People
Article Title: The Promise and Perils of Artificial Intelligence in Advancing Participatory Science and Health Equity in Public Health
News Publication Date: 14-Feb-2025
Web References: JMIR Public Health and Surveillance
References: DOI 10.2196/65699
Image Credits: Credit: JMIR Publications

Keywords: Artificial intelligence, Digital publishing, Public health, Health equity, Machine ethics