Background:  

The National Institutes of Health (NIH) is conducting planning activities to inform a potential Common Fund research program called Complement-Animal Research In Experimentation (Complement-ARIE). The program aims to advance the development, standardization, validation, and use of new methods and approaches that will more accurately model human biology, known as New Approach Methodologies (NAMs). These NAMs are intended to more closely model human biology and would complement, or in some cases, replace traditional models, transforming the way we do basic, translational, and clinical sciences.

As part of the Complement-ARIE strategic planning activities, the NIH Common Fund hosted three public listening sessions in October 2023. These virtual events brought together key representatives from multiple sectors, including industry and academic partners, non-government organization (NGO) representatives, and U.S. government and international partners, to gain insight unique to their fields regarding current opportunities and roadblocks in NAMs development. Approximately 1,100 people registered, and over 550 participated across all sessions. Interest groups were also invited to submit written input via email to the Complement-ARIE planning team.  

Participants provided a range of feedback centered around five main topics:

• Limitations of animal models
• Current limitations of NAMs
• Potential future applications of NAMs
• Building confidence and validation of NAMs  
• Enabling widespread adoption of NAMs

This report summarizes the comments provided during the listening sessions and via written feedback. These summaries represent the opinions and perspectives of the listening session participants, which do not necessarily reflect the perspectives of NIH or the federal government or the goals or structure of the potential Complement-ARIE program. For the purposes of these discussions, NAMs encompass a broad range of human-relevant complex in vitro systems, multi-scale computational models, high-throughput screening systems, in chemico (cell-free) assays, and other innovations to better understand human physiology and disease. While model organisms have been an important tool in biological research, the use of metazoan models will not be considered for Complement-ARIE, including but not limited to zebrafish, fruit flies, roundworms, water fleas, and frog embryos.

Participant Discussion of Limitations of Animal Models  

Participants discussed the limitations of translating findings from animal studies to humans for diseases involving interactions between complex systems such as the immune, nervous, and gastrointestinal systems, as well as host-microbiome interactions. This includes deficiencies in animal models for autoimmune diseases, neurodevelopmental and neurodegenerative diseases, cognitive and psychiatric disorders, pregnancy and reproductive mechanisms, cancer, respiratory diseases, and metabolic disorders.

Differences in physiology, immune response, genetic diversity, xenobiotic metabolism, toxicokinetics, pharmacokinetics, and lifespan also limit the applicability of animal models to human health and disease. For example, short-lived animal models are insufficient in replicating human aging mechanisms and chronic diseases. Animal models do not sufficiently replicate how diseases develop and manifest in humans, and there are few animal models for rare diseases. Additionally, the limited genetic diversity and variability in animal models do not adequately represent human population diversity.

Animal models also do not capture the dynamic social and environmental factors that affect human health. For example, lab animals are housed in sterile and homogenous environments that do not mirror the complex environments humans experience, including social determinants of health.

These limitations can hinder drug development, as many drugs that are successful in pre-clinical testing in animals fail in human clinical trials.  

Animals also do not adequately model subjective measures such as pain management, psychological disorders, or disability recovery. Furthermore, animals are often expensive to obtain and maintain in laboratory settings.

Representatives from NGOs also noted the negative toll working with laboratory animals has on biomedical workers’ mental health and well-being, particularly when the animals are in pain. Participants from the NGO and the U.S. government and international partners sessions recommended NIH conduct a systematic assessment to identify animal models with poor translatability to inform future funding decisions for NAMs.

Participant Discussion of Current Limitations of NAMs

Like animal models, NAMs have limitations in replicating population diversity, complex human biology and disease, and dynamic environments. Technology constraints, lack of standardization in NAM protocols, and availability of harmonized datasets are also challenges.  

NAMs are currently deficient in modeling age-dependent diseases and the progression of chronic disorders over time. This is because in vitro NAMs are typically developed using young cells (i.e., induced pluripotent cells), which do not mimic mature cells in the human body. NAMs that model neurological disorders and behavioral endpoints are also inadequate due to the lack of long-term data to model the chronic and cumulative effects of aging on cells in vitro.  

Participants discussed gaps in the ability of NAMs to accurately mimic diseases involving communication between multiple cell types, tissues, or organ systems in the body. However, a systems approach may not always be required, and considering the context of use can help minimize connectivity between tissues as a limitation.  

NAM platforms do not represent dynamic, human-relevant exposures, including changes in the internal environment (e.g., hormone fluctuations) and the synergistic effects of multiple chemical exposures and social stressors in humans. NAMs are also limited in their ability to model the metabolism and pharmacokinetics of drugs. In addition, current limitations for in vitro to in vivo extrapolation analyses must be resolved to improve the confidence in NAMs results. High cost and low-throughput are also barriers to the predictive capability and widespread adoption of current NAMs. The lack of standardization regarding data harvesting, NAMs development, and validation endpoints makes reproducibility across labs and research organizations a continuing challenge.

Current NAMs do not represent population diversity or variability, particularly for groups traditionally underrepresented in health research. Participants called for a focus on community engagement to recruit diverse donors. Limited access to data, tissue banks, and cell lines derived from a diverse set of donors to develop NAMs is also a constraint.  

Participants noted several types of NAMs had specific limitations. For example, in silico NAMs are currently constrained by a lack of high-quality training data and insufficiencies in the high-performance computing technology required to model complex diseases.  

Generally, participants spoke about how a lack of standardization in the NAMs field complicates inter-laboratory collaboration and acceptance by end users. Furthermore, data standards are needed to facilitate the sharing and harmonization of NAMs data. Increasing scientific confidence in NAMs is also an issue, and participants in all sessions spoke about the need for frameworks to standardize how NAMs are developed, how data is collected, and how validation is conducted. A database is needed to catalog existing NAMs that includes the context of use, results, validation status, and best practices for each NAM. The database can build upon or coordinate with existing NAMs databases as well. Participants provided examples of existing databases as a reference, including the Microphysiology Systems Database and the European Commission Tracking System for Alternative Methods Towards Regulatory Acceptance, which track NAMs and their validation status.  

Finally, representatives from federal agencies expressed that there are currently no guidelines for how to properly cite NAMs in publications, often making it difficult to discover what work is being done with NAMs in what areas.  

Participant Discussion of Potential Future Applications of NAMs

Despite current limitations, NAMs have a wide-ranging potential to accelerate biomedical research discoveries by improving our understanding of the mechanisms driving disease. NAMs can use existing datasets from large human population studies and patient-derived induced pluripotent stem cells (iPSCs) to facilitate research on disease initiation, progression, prevention, and treatment. They allow for improved modeling of a breadth of health outcomes, including rare diseases, neurological disorders, cancers, wound healing, and chronic illnesses across the lifespan. The use of digital twins and virtual patient models are promising approaches that could achieve these advancements, as noted by representatives from the U.S. government, international groups, and NGOs.  

Participants in all sessions posited that in silico and in vitro approaches can facilitate the development of safer, more effective drugs. NAMs can be used to predict adverse clinical trial outcomes, test for potential adverse drug interactions, and generate more accurate drug metabolism profiles. Patient-derived iPSCs can enable a precision medicine approach, thus providing patients with personalized disease prevention and treatment options.

Additionally, NAMs can be used to conduct large-scale, high-throughput testing of chemicals to facilitate complex risk assessments, such as for large groups of chemicals like PFAS or for multi-dose exposures. Participants also expressed that NAMs have the potential to facilitate modeling and predictions for rapid response situations, such as infectious disease outbreaks.  

Participants noted that integrating data across various NAM platforms, anchoring NAMs in the whole-body context, and incorporating genetic and other types of diversity into NAMs will be critical to advancing the field.

Participant Discussion of Building Confidence and Validation of NAMs  

Barriers to validation and widespread acceptance of NAMs include uncertainty around validation requirements and processes, low confidence in NAMs data by regulators and end-users, lack of familiarity with NAMs data, and lack of support for validation studies.  

To build confidence in NAMs, participants called for a government-led effort to educate all partners on how to interpret NAMs data and the methods’ usefulness and limitations. Participants from all sessions agreed that creating opportunities for regulators, developers, and end-users to collaborate early and often is necessary to facilitate NAMs development, validation, and adoption. Furthermore, participants across sessions called for funding initiatives to support validation studies and inter-laboratory comparisons of NAMs assays.  

Representatives from academia and NGOs noted the need to set reasonable regulatory expectations for NAMs. It is critical that NAMs demonstrate relevance to human biology by reflecting key events along adverse outcome pathways. Some participants also noted the importance of validating NAMs against existing in vivo data to facilitate regulatory acceptance and uptake by end users.  

The development of a framework outlining validation requirements and steps is needed. This framework should be developed collaboratively, focus on the context of use, and be flexible enough to allow for cases when comparison to animal data is not possible or appropriate.  

Participants in all sessions recommended that the NIH invest in infrastructure to support NAMs validation, including tissue banks, data-sharing infrastructure, and a registry of NAM methods and data that have been accepted for regulatory purposes and their context of use.  

Participant Discussion of Enabling Widespread Adoption of NAMs

Achieving widespread adoption of NAMs will require collaboration among industry, academia, government, and regulatory agencies. These groups should work together to establish shared goals, achieve standardization, and share data and resources. Cooperation between NAMs developers and end users is also critical. Representatives from academia and industry noted that public-private partnerships for data sharing will be key to accelerating NAMs development and reducing end users’ uncertainty around NAMs.  

The preference for animal-based methods is another barrier to the widespread adoption of NAMs. Participants recommended an educational initiative aimed at grant reviewers, journal publishers, regulators, and end users to help reduce animal methods bias, demonstrate NAMs relevance to human biology, and build confidence in the methods. Standardization of NAMs data, relevant endpoints, and validation criteria is also needed to increase adoption and acceptance.  

Training scientists, particularly early-stage investigators, in NAMs is needed for adoption and continued growth in the field. Training should focus on NAMs development, use, limitations, and oversight. Participants recommended institutional training programs, center programs, and administrative supplements as approaches to train researchers while promoting collaboration.  

Access to costly NAMs infrastructure and disparate data is a major barrier to uptake. Creating a centralized tissue bank and data repository and leveraging existing initiatives, like the NIH All of Us Program, can help researchers access the resources needed to transition from traditional studies to NAMs research. Participants across sessions recommended the establishment of a center program to provide researchers access to shared core facilities and the interdisciplinary expertise necessary to stimulate NAMs adoption and progress.  

Collaborations and partnerships are also key to developing NAMs and the infrastructure and expertise needed to support the field. Interdisciplinary collaborations should bring together experts in materials science, computational science, toxicology, clinical research, and drug discovery. International partnerships will be key to building and supporting NAMs infrastructure, such as biobanking repositories, AI databases, and a data ecosystem. Funding opportunities are needed to incentivize scientists and institutions to undertake the substantial financial investment necessary to upgrade infrastructure to support NAMs research. Additionally, participants in the NGOs session noted that funding is needed to increase access to NAMs equipment and resources for Contract Research Organizations and researchers working outside the U.S. or Europe.  

Conclusion of Participant Views

NAMs offer many advantages over traditional models for replicating human biology and disease. Collaboration across interest groups and investment in training, data sharing, and research infrastructure is needed to advance the development, standardization, validation, and use of NAMs. NIH, in partnership with other U.S. government and international agencies, has a leadership role in planning, coordinating, and funding the initiatives that will encourage scientists to use NAMs and accelerate biomedical discoveries that can improve human health.

On Oct. 19-20, 2023, representatives from eight federal agencies convened in Washington, D.C., for the Complement Animal Research In Experimentation (Complement-ARIE) Inter-Agency Retreat, hosted by the National Institutes of Health (NIH) Common Fund, to discuss goals and priorities for advancing new approach methodologies (NAMs). The aim was to lay the foundation for a new Common Fund program to catalyze the development, standardization, validation, and use of human-based NAMs to transform biomedical research and improve predictions of human health and disease.

The proposed Common Fund program would:

• Develop human-based NAMs that can better model health and disease across diverse populations.
• Address the limitations of biological insights that current models provide.
• Complement traditional models to make research more efficient and effective.
• Support regulatory decisions to improve human health.

Through presentations, panel discussions, and breakout sessions, agency representatives in attendance highlighted the various NAMs underway in their respective organizations and fields. These approaches include organs-on-chips, virtual models, 3D cell cultures, organoids, stem cell models, computational methods, and more. As stated by retreat participants, scientific areas that would benefit from NAMs include cancer research, clinical trials, chronic diseases, aging, adaptive immunity, neuroscience, and precision health, among others.

Challenges and Priorities

The retreat underscored several priorities that will enable broader acceptance and impact of NAMs. Agency representatives expressed that clear standards are needed across all aspects of NAMs, from data generation methodologies to validation frameworks to reproducibility efforts. Robust standards will encourage regulatory approval and use of NAMs. According to attendees, regulatory agencies should also require that NAM data adhere to FAIR (findable, accessible, interoperable, and reusable) principles.

NAM data should be freely available, said agency partners during moderated conversations. A database as well as harmonized data-sharing networks would facilitate the development of new NAMs and highlight gaps where NAMs are unavailable, as well as areas where traditional models are not predictive of human responses.

Advancing NAMs requires bringing together diverse data, technologies, disciplines, and collaborators.

Validation of NAMs remains a significant challenge.

Training and educating early-career scientists is vital to increasing NAM adoption. There is also a need to train current and future scientists and regulators, as well as the public, on the uses and limitations of NAMs. Such efforts would increase regulatory understanding of NAMs and encourage widespread acceptance and adoption of NAMs in laboratory spaces.

Transparent communication is needed to develop public understanding of and trust in NAMs. 
The most effective, resource-efficient use of NAMs is to complement traditional models. Although creating NAMs to replace traditional models might be applicable in some situations, the goal is to create NAMs that fill knowledge gaps that current models cannot address.

Collaborative Opportunities

Partnerships between industry, agencies, and academia will be necessary to develop shared standards and processes, which in turn will aid in the adoption of NAMs Retreat attendees also discussed focusing on developing NAMs that address worldwide issues.

Industry partnerships will be essential to the continued development and eventual acceptance of NAMs.

Collaborations can also fulfill certain pressing issues. High-priority NAMs — many of which are needed to alleviate longstanding gaps in knowledge — were identified for future development and validation.

Other collaborative opportunities for NAM integration include therapeutic drug development and safety and efficacy tests for medical products.

Insights From the Breakout Sessions

Participants were separated into two groups for two breakout sessions. Both groups discussed regulatory and scientific needs for NAMs and partnership opportunities. Several high-level themes that emerged from the session follow:

• Regulatory needs and opportunities for partnership
• A database of accepted and validated NAMs could inform NAM development and regulation by helping identify areas where traditional models are not sufficient or not available.
• The amount of data available for NAM development and validation must be increased and be freely available, harmonizable, and reusable.
• Additionally, standardized, established frameworks should ensure data quality, reproducibility, and utility.
• Context of use is critical when considering regulatory acceptance.
• Communication and confidence-building about NAMs will increase acceptance across a range of collaborators. Establishing which NAMs are useful for which settings will be important in this process.
• Scientific needs for NAMs and opportunities for partnership
• Training the next generation of scientists on interdisciplinary research for developing and using NAMs is a critical need.
• Incentives are needed to encourage academics and industry to use NAMs. Current models must be improved to increase clinical relevance, predictive power, and diversity.
• Standards, benchmarks, and shared data infrastructure should guide NAM development. Interoperability should be prioritized to facilitate harmonization, which will include addressing data concept consistency and data access issues.

The Path Ahead

Retreat attendees demonstrated that demand for NAMs is strong, but the landscape is underdeveloped. Collaboration between interested parties is essential for future NAM development, validation, and dissemination. Although various federal agencies have made initial efforts to integrate NAMs, a more coordinated effort is necessary. A Common Fund program centered on NAMs could lay the foundation for global acceptance of these biomedical tools.

Team members: Zev Gartner, University of California San Francisco (team captain); Ophir Klein, University of California San Francisco, Cedars-Sinai; Faranak Fattahi, University of California San Francisco.

Models of the human gut will revolutionize our ability to test drugs and study disease but existing models make inherent compromises with respect to physiological relevance, complex functions, and reproducibility that limit their ultimate applications. We propose a new approach to direct the formation of gastrointestinal tissue that will include multiple integrated organ systems capable of involuntary gut muscle contractions and digestive functions. To build this model we use a process called 4D tissue fabrication. We use a next generation bioprinter to specify the initial 3D coordinates of dense slurries of stem cells in support baths optimized for their self-organization—a process mimicking their natural development. The same bioprinter then actively guides self-organization into more complex and functional tissues over time (i.e. the 4th dimension). We have demonstrated proof of principle of this approach to build complex and perfusable models of intestinal tissue and vasculature. Here, we propose to expand this approach to make models with orders of magnitude more structural complexity and in the most physiologically relevant context. The platform is generalizable to a variety of other tissues.

Team members: Riccardo Barrile, University of Cincinnati (team captain), Ryan White, University of Cincinnati; Wayne Poon, NeuCyte.

Brain diseases like tumors and age-related conditions are one of the main causes of disability and a common cause of death worldwide. The challenge is that we lack good ways to study and treat these diseases. Current methods, often involving testing on animals, don't always give results that apply to humans because the human brain has unique characteristics. Our project offers a new approach using special cells called stem cells to assemble miniaturized models of the brain (minibrains) including a protective barrier of the brain called the blood-brain barrier. This barrier normally shields the brain from harmful substances in the blood. But in people with brain diseases, this barrier often doesn't work properly, making the disease worse in ways we don't completely understand. Think of our model as a personalized "Avatar" of a patient's brain, capturing important aspects of their condition to enable personalized treatments. We're focusing on Alzheimer's disease and using cells from patients. These cells are grown in a platform with multiple sensors that monitor brain cell activity in real-time, helping us gather important data in a consistent way. Our goal is to create a model that closely imitates human conditions, giving us better insights and more accurate predictions on how patients could respond to treatments. This could be a game-changer in understanding and treating brain diseases.

Team members: Ken Chen (team captain), University of Texas MD Anderson Cancer Center; Yujia Wang, University of Texas MD Anderson Cancer Center; Stefano Casarin, Houston Methodist Research Institute; May Dahar, University of Texas MD Anderson Cancer Center; Vakul Mohanty, Univ. of Texas MD Anderson Cancer Center.

Adoptive cellular therapies (ACT) treat cancer using immune cells harvested from patients or healthy donors/sources. Despite their tremendous potential, predicting benefits is difficult due to multifaceted complexity at the molecular, cellular and systems levels. Numerous experiments using cells, animals and patients are required to study the complexity, before an effective treatment can be developed. Similar to other complex domains, computer models such as agent-based models (ABM) can provide cost-efficient, complementary, and meaningful ways to simulate the tumor environment and predict treatment outcomes. Leveraging our strengths in ACT development and trials, single-cell omics technologies, and computational modeling, we plan to incorporate molecular profiles of tumor cells and immune cells into ABMACT, an agent-based model for adoptive cell therapies to study treatment responses of B cell lymphoma by Chimeric Antigen Receptor Natural Killer (CAR-NK) cells, a type of engineered immune cells under various conditions. With ABMACT, we aim to improve adoptive cellular therapy product design and treatment planning for personalized cancer immunotherapies.

Team members: Weiqiang Chen, New York University (team captain); Chao Ma, New York University; Lunan Liu, New York University

CAR T-cell immunotherapy that uses and enhances patients’ own T-cells to fight cancer has emerged as an innovative method for treating various types of cancers. Despite its remarkable success in treating blood cancers, CAR T-cell therapy is less effective in solid tumors such as pancreatic cancer, which is largely attributed to the tumor heterogeneity and its “immune cold” tumor microenvironment. Lack of reliable clinical method to rapidly and accurately assess the potency of patient-derived CAR T-cell products before administration thus predict patient response poses a major clinical challenge. To develop improved and personalized immunotherapies, we will invent a novel patient-derived “pancreatic cancer organoids-on-a-chip” microphysiological system to reconstruct patient-specific tumor microenvironment on a chip. We will use this system as an integrated precision medicine tool for an accurate evaluation of CAR T-cell therapy efficiency in individual patients. Such a hybrid “in silico-in vitro” modeling system integrates in vitro organoids, organ-on-a-chip technologies, in silico modeling and in vivo clinical data to better emulate the in vivo patient pathology and heterogeneity for CAR T-cell therapy screening. Our solution can be implemented into clinical practice and provide a new paradigm for “clinical trials on a chip” that leads to the development of personalized CAR T-cell immunotherapy strategies for cancer patients.

Team members: X. Lucas Lu, University of Delaware (team captain); Michael Axe, University of Delaware; Zugui Zhang, University of Delaware; Joseph Fox, University of Delaware.

Our New Approach Methodology (NAM) is based on click chemistry and bioorthogonal reactions, powerful techniques that received the Nobel Prize in Chemistry in 2022. This method measures how quickly cells and tissues can create new proteins, making it highly useful for medical research and drug discovery.
When compared to traditional methods, our NAM is more reliable, accurate, flexible, easier to use, cost-effective, and safer. It can significantly enhance our understanding of human health and diseases, reducing the reliance on animal and human testing. This method's global applicability ensures that research findings are trustworthy and repeatable. 

We are using this versatile method to test over 3,000 FDA-approved drugs on human knee cartilage samples, evaluating their potential for treating osteoarthritis. We'll assess the effectiveness of these drugs under both normal and inflammatory conditions. We are employing machine learning and artificial intelligence (AI) algorithms to analyze the data. 

Beyond the realm of drug discovery, our NAM has the potential to revolutionize cancer treatment by enabling the creation of personalized therapies from tumor biopsies. It is also highly effective for fabricating artificial tissues in laboratories. Overall, our NAM could significantly transform medical research, making it more inclusive, effective, and globally impactful.

Team members: Alexander Tropsha, Predictive LLC. (team captain); Eugene Muratov, University of North Carolina at Chapel Hill (UNC), Predictive, LLC.; Kevin Causey, Predictive LLC.; Greg Sokolsky, Predictive LLC.; Ricardo Tieghi, UNC.

The proposed research addresses a critical gap in understanding the safety of medications for pregnant women, where 80% use prescription drugs, yet drug safety is grossly understudied. Clinical trials rarely include pregnant women, leaving gaps in knowledge about potential fetal toxicity. The project's aims include creating a knowledge base on developmental toxicants, or agents that cause toxicity, building molecular models, constructing a knowledge graph, and developing a web portal for real-time toxicity predictions. The significance lies in the unmet need for testing toxicity in pregnant populations, using artificial intelligence (AI)-driven technologies to enhance drug safety during pregnancy. Predictive LLC proposes innovative approaches to identify developmental toxicant associations, emphasizing the creation of a novel developmental toxicity knowledge graph. The team's prior research demonstrates rigor, including developing approaches like Chemotext and ROBOKOP. The research introduces novel computational methodologies, leveraging AI to accelerate drug discovery for pregnant populations. The cost-effectiveness of computational approaches is emphasized, highlighting potential cost and time savings compared to traditional drug discovery. The unique capabilities include establishing a comprehensive knowledge graph and machine-learning models to infer and test novel associations, addressing a critical need in pharmaceutical research.

Team members: Thomas Hartung, Johns Hopkins University (JHU) Center for Alternatives to Animal Testing (CAAT) (team captain); Tom Luechtefeld, JHU CAAT; Alexandra Maertens, JHU CAAT.

New testing methods to replace animal studies often fail because validating them takes too long and costs too much money. Our idea uses artificial intelligence (AI) to plan validation studies in a smarter way. The AI software suggests the best and most representative reference chemicals to test. It runs computer simulations to figure out exactly how much data is needed to get reliable results. The AI also checks databases of past studies so new studies don't repeat work already done. Additionally, it provides training for researchers on how to correctly use and interpret the new testing methods. By making the validation process faster, cheaper, and more rigorous, our AI platform opens the door for more innovative non-animal testing approaches to be put into practice. This will accelerate the testing of medicines and chemicals for safety, unlocking the potential of human-focused research methods to improve human health

Team members: Shantanu Singh, Broad Institute of MIT and Harvard (team captain); Constance Mitchell, Health and Environmental Sciences Institute (HESI); Christine Crute, HESI; David Rouquie, Bayer; Andreas Bender, University of Cambridge; Anne Carpenter, Broad Institute of MIT and Harvard; Srijit Seal, Broad Institute of MIT and Harvard.

It is currently very difficult to predict whether a potential new medicine or agricultural chemical is safe even after animal testing. Often toxicity is only discovered later - during clinical trial testing in people or broader use. This causes tremendous harm and these failures drive up the costs of new, useful chemicals. As part of a Consortium, “Omics for Assessing Signatures for Integrated Safety” (OASIS), we are pioneering a new approach to predicting a chemical’s liver toxicity, via a first-of-its-kind strategy. OASIS aims to capture a broad spectrum of information from liver cells treated with compounds using assays that simultaneously measure more than a thousand cell responses, including Cell Painting, mRNA levels, and protein levels to predict compound liver toxicity. We will identify which of these responses when combined using machine learning, can best predict outcomes in the whole organism. We bring together scientists from 14 pharmaceutical and agrochemical companies and 6 technology companies. If successful, the Cell Painting assay, coupled with molecular-omic data, will reduce animal testing and allow earlier, more accurate, and less expensive liver safety testing of candidate compounds across two industries. As well, our datasets will be made public to serve as a foundation for future advancements.

Team members: Brian Johnson, Michigan State University (MSU) (team captain); Sudin Bhattacharya, MSU Dept of Biomedical Engineering (BME) and PHM & TOX; Jacob Reynolds, MSU Dept BME.

The human body is made up of cells that communicate with each other to develop and function properly. If this signaling is disrupted during development, birth defects such as orofacial clefts can occur. Since the process of cell to cell signaling is complex and not well understood the cause of most structural birth defects remains unknown. To address this challenge, we have created a microphysiological model that mimics the interactions between cells during orofacial development. Our model is designed to study normal and abnormal development, such as clefting. It incorporates important tissue structure and function and is manufactured within a microplate using computer numerical control (CNC) micromachining, making it easy to adopt by other labs. By adding human-derived cranial neural crest cells and familial risk variants for clefting, we can capture population variability and create a model predisposed to clefting. We will perform single cell sequencing of the mesenchyme, neural crest, and epithelium to gain insights into cell to cell interactions. This will help us understand the causes of birth defects and develop ways to prevent them in the future. This work aims to increase understanding of the cell to cell communication that drives development so birth defects can be better understood and prevented. 

Team members: Lei Yin (team captain), Reprotox Biotech LLC.

There's growing concern about the impact of chemicals in our environment and drugs on male reproductive health. These chemicals or drugs can potentially harm sperm production and fertility. However, testing for these effects is currently expensive, time-consuming, and often relies on animal models which may not perfectly reflect human reactions. This research proposes a pioneering approach to tackle this challenge. The goal is to develop a new method for testing chemicals that directly uses human cells, grown in the lab. This method aims to create tiny models of human testes, using stem cells and other supporting cells involved in sperm production. These "Mini-Testis" will then be exposed to different chemicals to observe how they affect the cells and reproductive processes. This novel model will provide faster and more affordable testing as compared to animal testing. Also, this human mini-testis model using human cells will provide more accurate predictions of how chemicals might affect people, as it directly reflects human biology. Studying these "human Mini-Testis" could also lead to a deeper understanding of how chemicals cause reproductive harm. This knowledge could pave the way for developing new treatments for male infertility and other reproductive health issues. 

Overall, this research has the potential to transform how we test for male reproductive toxicity, leading to more efficient, accurate, and ethical testing methods.

Team members: Carlos Ramon Ponce, Harvard Medical School (team captain); Alireza A. Dehaqani, Harvard Medical School; Antonio Montanaro, Harvard Medical School; Giordano Ramos-Traslosheros, Harvard Medical School; Olivia Rose, Washington University in St. Louis, Visiting Student of Harvard Medical School; Binxu Wang, Harvard University.

Deep learning models have achieved extraordinary accuracy in image recognition, reaching and sometimes exceeding human levels. However, their internal analyses are largely unknown, making it difficult to know if these systems could malfunction and if so, how to fix them. This project aims to demystify these models, using principles of visual neuroscience. Our library, ATHENA-N (Analyzing The Hidden ENcoding in Artificial Neural Networks), will methodically analyze deep neural networks, illuminating their learned representational structure and information processing methods. 

ATHENA-N includes several modules to investigate how these networks represent and process visual information. The goal is to establish connections between artificial and biological neural encoding, enhancing our understanding of both fields. 

ATHENA-N will set the path to new biological discoveries, identifying new potential functional neurons present in visual cortex, and helping us define how “brain-like” some artificial intelligence (AI) models are. This will be instrumental in developing AI systems that are transparent, trustworthy, and reliable for real-world applications. Moreover, this project is poised to generate numerous research questions and hypotheses, fostering cross-disciplinary collaborations. 

Team members: Wei Tan, University of Colorado at Boulder (team captain); Chuangqi Wang, University of Colorado at Denver.

Cardiovascular diseases are the leading cause of morbidity and mortality in the United States and around the world. A major form of the diseases is coronary heart disease or atherosclerosis. There are constant needs for a deeper understanding of the pathways that lead to this disease and how to better prevent and treat it with new therapies. Our multi-disciplinary team aims to leverage the power of artificial intelligence, multi-functional biomatrix engineering, and high-throughput screening tool to create human-based disease model. If successful, such a model may facilitate mechanistic and therapeutic investigations of vascular diseases, specifically atherosclerosis. Additionally, the integrated platform model would allow one to predict and validate the impacts of diseased tissue properties on the efficacy and potency of cardiovascular treatments. The platform technology also has the potential to extend well beyond the optimization of cardiovascular treatments, expediting the therapeutic discovery for other types of diseases. The innovations of the proposed model include its capacity of offering (a) realistic, interactive parametric explorations for a multitude of inputs (e.g., various treatments, biophysical and biochemical properties of diseased tissues) and outputs (e.g., disease stage-characteristic behaviors such as inflammation, proliferation, function loss), and (b) highly efficient and economic approach towards translation and commercialization of new therapies.

Team members: Thomas Luechtefeld (team captain), Insilica LLC.; Zakariyya Mughal, Insilica LLC.; Thomas Hartung, Johns Hopkins University.

NAMKG stands for New Approach Methodology - Knowledge Graph. It's like a big online library where researchers can find and share information about new scientific tests used in labs. These tests are called New Approach Methodologies (NAMs). Researchers can put details about their NAMs in NAMKG, making it easier for others to find and understand them. 

NAMKG uses artificial intelligence (AI) to keep this information up-to-date. It's helpful for scientists working on new drugs and for people making policies about science and health. NAMKG brings together researchers, industry experts, and regulatory people to make sure scientific methods are good and reliable. This helps science work better and makes it easier for everyone to understand and use scientific discoveries. 

NAMKG also allows technology developers to more easily access information about new scientific methods. This means that software developers can make their applications better by pulling in this public data. NAMKG even allows software developers to add their own software tools for the system. Once a tool is added to the system, NAMKG allows its AI agents to leverage those tools in conversations with users, so you can ask a question like "Is my chemical hazardous?" and NAMKG will run all the automated tests it can to answer your question. 

Ultimately, NAMKG is a resource that integrates information about new scientific methods, and makes it easier for people to access, learn about, and leverage those methods for their own work.

Team members: David Kaplan, Tufts University (team captain); Liam Harrington Power, Tufts University; Olivia Foster, Tufts University; Michael Lovett, Tufts University.

Historically, animals have been used to test new cosmetics, drugs, and other chemicals to understand their effects and evaluate safety concerns, but animals are not always good options to mimic human body function, physiology, and behavior. Many countries have banned the use of animals for these purposes, requiring other options for safety and efficacy testing. Lab developed models of three-dimensional tissue systems with human cells have been shown to be good models of human tissues and can be used to predict how chemicals and pharmaceuticals may interact with the body. Tissue models of human skin exist, but they are simplified by focusing only on the top layers of skin and do not consider deeper layers that contain relevant features for testing, such as immune system and nerve components, which are critical to assess toxicity and irritation from chemicals. More complex models that include the full thickness of human skin and its various features such as immune cells, blood vessels, and nerves would provide more accurate comparisons as testing platforms relative to the human body than current, more simplified skin models are able to provide. We are proposing a new model system that includes all three layers of human skin and many relevant features, including immune cells, blood vessels, and nerves, to generate a more accurate system to test new compounds for safety and efficacy outcomes with human relevance. 

Team members: Hang Lin, University of Pittsburgh (team captain); Meagan Makarczyk, University of Pittsburgh; Johannes Plate, University of Pittsburgh; Daniel Kaplan, University of Pittsburgh; Claudette M. St. Croix, University of Pittsburgh; Michael Gold, University of Pittsburgh

Osteoarthritis (OA) is a painful and debilitating disease that is also highly prevalent. For example, 27 million US citizens suffer from osteoarthritis, including ~25% of those >50 years of age. Pain is one of the primary reasons that osteoarthritis patients seek medical attention, yet there are no consistently effective treatments for osteoarthritis pain that are not associated with potentially fatal side effects. 

Limited progress has been made in the study of current models of OA pain, such as animals. The team recently pioneered the development of an in vitro microphysiological tissue chip, which contains four critical elements of the knee joint (cartilage, bone, synovium, and fat pad) engineered from human cells (patent granted). Moreover, human sensory neurons (the cells that generate pain activities) were included to form pain-enabled knee joint chips (Neu-miniJoint, patent pending). 

Here, we aim to further increase the complexity and clinical relevance of the Neu-miniJoint by including immune cells (the cells that provide molecules to induce pain) in the system. OA-like conditions will be induced through mechanical injury. The pain-associated activities in the neurons will be examined by established methods in the team. 

Successful generation of this new model containing neural and immune cells (ImNeu-miniJoint) will not only identify factors responsible for OA pain but also enable the development of truly personalized osteoarthritis pain medications.

Team members: Lena Smirnova, Johns Hopkins University (team captain); Thomas Hartung, Johns Hopkins University; David Gracias, Johns Hopkins University; Brian Caffo,  Johns Hopkins University; Itzy Erin Morales Pantoja, Johns Hopkins University; Dowlette Alam El Din, Johns Hopkins University; Lomax Boyd, Johns Hopkins University.

The human brain is amazingly complex, but studying it is challenging. Experiments often use animals, especially monkeys and apes. But advanced cell models using human induced pluripotent stem cells now allow us to grow brain organoids (miniature 3D cultures of brain cells) and study their functions in the lab dish.

We propose combining organoids with artificial intelligence (AI) for "organoid intelligence" to create a learning-in-a-dish model. We will give input signals and track output to see how brain organoids react and change behavior over time, like simple learning. This will help us understand how the brain works without using animals.

Our brain organoids will start simple but add more brain cell types/regions and connections over time. We will give electrical and chemical signals and track responses with electronics to study memory and information processing. AI will analyze the complex data so we can understand and improve the brain organoids.

This research will let scientists study real human brain cells in action! We hope it advances brain science by showing how brains compute. We have to carefully address ethical concerns, but organoid intelligence has huge potential for discovery about cognition without distressing animals like monkeys. Exciting innovations could lead to better brain-like computers and therapies too!

Team members: Deepak Rao, Brigham and Women's Hospital (team captain); Kathryne Marks, Brigham and Women's Hospital; Kevin Wei, Brigham and Women's Hospital.

Autoimmune diseases affect over 20 million Americans and cause disease in almost every organ system. Current drugs to treat these diseases provide blunt tools that are broadly immunosuppressive, affecting both pathologic and protective immune responses. There is a critical need for new, more selective therapies and improved understanding of how to use therapies. New methodologies leveraging human cells and tissues are needed to improve our ability to interrogate new therapies and select the right drug for individuals.  Rheumatoid arthritis (RA), a common, severe autoimmune disease that targets joints, provides an ideal context in which to develop new methodologies since multiple therapies are approved to treat RA, yet there are no robust biomarkers to predict which drug a patient will respond to. We propose to use synovial organoids, built using cells from RA patients, to model the pathologic autoimmune response in joints. We propose that organoid systems will recapitulate relevant cell interactions and capture the diversity in tissue injury patterns observed in RA patients. We expect that in vitro organoids of RA joint tissue will enable rapid, more faithful interrogation of effects of new therapeutics on the inflammatory environment in the target tissue and will define the cellular pathways most affected by individual therapies. This organoid methodology can be readily extended to dissect inflammatory pathways in other autoimmune and inflammatory diseases. 

Team members: Rebecca Pompano, The Rector and Visitors of the University of Virginia (team captain); Chance John Luckey, The Rector and Visitors of the University of Virginia; Jennifer Munson, Fralin Biomedical Research Institute at Virginia Tech; Evangelia Bellas, Temple University; Aarthi Narayanan, George Mason University.

Vaccination and immunotherapies are critical tools for public health but remain challenging to develop, often requiring decades to reach clinical use. Progress is especially slow for brain or immune diseases, as these tissues are largely inaccessible for study in humans and animal models are insufficient to replicate their complexity or the impacts of human population diversity. Therefore, we envision a groundbreaking, in vitro model to simulate the interactions between the human brain and immune system, while beginning to account for diversity of sex, age, race, and changes in metabolic state due to leanness or obesity. Our model starts with cells originally from human donors that are cultured to form tiny replicas of the brain, lymph node, and adipose (fat) tissue. The three replicas then integrated into a single housing, smaller than a credit card, and connected via two intersecting loops of continuously recirculating fluid that represent the intersection between the brain and the rest of the body.  We will add more cell types and functions to this system until it replicates specific aspects of infection with viruses that infect the human brain, as well as the ability of clinical vaccine candidates to protect (or not) against those viruses. Success will yield a platform for studying and treating neuro-immune infections including influenza, SARS-CoV-2, and potential biothreats, along with neurodegenerative conditions like Alzheimer’s, multiple sclerosis, and brain cancer.

Team members: Ramkumar Menon, University of Texas Medical Branch (UTMB) (team captain); Arum Han, Texas A&M University; Lauren Richardson, UTMB; Ananth Kumar Kammala, UTMB; Moumita Chakraborty, UTMB.

Human pregnancy and giving birth are fascinating but perplexing phenomena to understand, as two independent biological and physiological systems (fetus and the mother) co-exist and must be simultaneously considered. Pregnancy associated complications like premature birth, miscarriages, and stillbirths are challenging to study, and medical interventions are difficult as researchers do not have good models to predict drug efficacy under these circumstances. There is a lack of knowledge of whether therapeutics are beneficial to the fetus (or harmful to in-utero fetal growth) and regulatory agencies (FDA) do not approve drug use during pregnancy. To better understand pregnancy biology, the birthing process (normal and abnormal), and to test drugs during pregnancy, we will create the entire human Pregnancy-On-a-Chip (POC) (a miniaturized device on a plastic slide) using cells derived from both the mother and the fetus after term delivery of a healthy live baby. POC recreates human pregnancy in the lab with all fetomaternal uterine components. This device will allow us to study cellular mechanisms that maintain normal pregnancy, labor, and delivery and recreate situations in which adverse pregnancy occurs. This understanding and model will enable us to test drugs that can mitigate pregnancy risk. Preclinical drug trial data using POC will accelerate clinical trials during pregnancy and save maternal and fetal lives. Further, POC will reduce the use of unreliable animal models as they don’t mimic human pregnancy.

Team members: Huaxiao 'Adam' Yang, University of North Texas; Gautham Mohanraj, University of North Texas; Marcel El-Mokahal, University of North Texas.

Multi-scaled engineered heart constructs (MEHCs) are revolutionary multiscale tissue cultures derived from human stem cells. At a microscopic level, these models can imitate intricate heart functions and structures, enabling groundbreaking studies of human heart physiology and diseases to complement animal models. With immense potential, the exploration of MEHCs promises to transform disease modeling and drug discovery.

At the center of this research is the novel software: Organalysis. Armed with formidable processing and analytical capabilities, this technology empowers researchers at the frontline of innovation by enriching MEHCs images and gleaning quantitative data about the function of the MEHCs. As such, Organalysis can transform biomedical research, opening the door to AI and ML-based techniques.

After developing this software, we then aimed to create a predictive model that detects cardiovascular disease patterns using data from MEHCs. Building off of the established framework for effective multiscale tissue modeling, this approach elevates in vitro research and enables customized predictive modeling tailored to patients. 

Overall, the integration of these technologies could catalyze pioneering discoveries that accelerate the clinical workflow from diagnosis to optimized, personalized treatment.