We’re thrilled to share that Deep Origin and our partners in the Pharmacological Research and Evaluation through Digital Integration and Clinical Trial Simulation (PREDICTS) consortium have been awarded an up to $31.7 million contract from the Advanced Research Projects Agency for Health (ARPA-H) Computational ADME-Tox and Physiology Analysis for Safer Therapeutics (CATALYST) program! CATALYST is led by ARPA-H Health Science Futures Mission Office Acting Deputy Director Andy Kilianski, Ph.D.
Through the ARPA-H CATALYST program, we are developing in silico and ex vivo systems to better predict drug safety and dosing, and reduce the use of slow, costly, and lower-accuracy animal studies before clinical trials ever begin.
Our ultimate goal: a future where first-in-human studies rely primarily on powerful in silico safety data -- eliminating the need for animal testing.
The problem
Animal testing has contributed tremendously to drug safety, but we may be hitting a fundamental limit in what it can tell us about the human body.
Most drugs that pass early animal studies -- designed to test absorption, distribution, metabolism, excretion, and toxicity (ADME-Tox) - still fail in clinical trials. In many cases, the human system simply behaves differently. These animal models are slow (adding 4-5 years to development), expensive ($2-4M per molecule), and have only 43-63% concordance with human toxicity outcomes. The result is clinical trial failure costs that can reach $50-60 billion annually in oncology alone [1-4]. That’s just the drugs that fail in the clinic, as well. We don't know how many drugs that may have worked safely in humans were axed because of a spurious safety signal in animal studies or biology that doesn't translate.
Researchers have been hard at work developing ex vivo technologies (like organoids and organ-on-a-chip systems) and in silico models to replace animal use, but both face major hurdles.
- Ex vivo models trade off complexity and throughput. You can test many compounds in simple models quickly-or a few in complex systems slowly.
- In silico models, while faster and scalable, are often piecemeal, trained on fragmented datasets, and lack integration between biological systems. Many remain unvalidated, opaque in mechanism, or computationally intensive to scale.
- Both face regulatory uncertainty -- it’s only recently becoming clear when and how the FDA might accept data from these models.
Adopting more predictive preclinical models faces challenges such as usability issues, job security fears, and risks of inaccurate results [5-7]. An ideal system would not only replace slow and inefficient studies with something more predictive but would provide information throughout preclinical drug development to create safer drugs.
Our approach: PREDICTS
The PREDICTS Consortium, led by Deep Origin, brings together experts in computational chemistry, in vitro and ex vivo assays, pharmacology, and AI to tackle these challenges head-on.
Consortium partners include:
- Deep Origin (San Francisco, CA / Yerevan, Armenia)
- Ginkgo Bioworks (Boston, MA)
- ImmVue Therapeutics (Toronto, Canada / Boston, MA)
- MIDO LLC (Cincinnati, OH)
- Netrias (Washington, DC)
- Sanford Burnham Prebys (La Jolla, CA)
- Tessel Biosciences (Cambridge, MA)
The consortium includes more than 100 specialists spanning computational biology, toxicology, and drug development.
Additional support will be provided by two particularly renowned consultants: Marc Birtwistle, PhD -- expert in computational and experimental cancer cell biology, mechanistic models of biochemical cellular responses to drugs - and Jan Hasenauer, PhD - who has vast experience in computational systems and cell biology including simulation software development.
Om Therapeutics will play an integral role in the wider project by empirically screening entire proteomes against hundreds of millions of small molecules to generate protein binding data at scale for ML powered discovery.
The combined team has in-depth experience in ex vivo experimentation with in silico simulation to build explainable safety models across a range of organs, tissues, and toxicology types. Additionally, we have a strong track record of delivering on government contracts with agencies like BARDA, DARPA, and NIH.
Why now
The scientific community has long recognized the limits of animal testing. It’s slow, expensive, and often misleading. Over the past decade, R&D spending has ballooned while the number of approved drugs per year has remained flat [6]. The FDA has acknowledged that the future of drug development can’t rest on animal models alone. In 2022, the FDA Modernization Act removed the requirement for animal studies in drug development, allowing FDA to waive animal studies at their discretion. In 2025, the FDA released a roadmap to reducing preclinical animal testing that eliminated the need for animal studies for certain biologics. And in collaboration with other agencies like the NIH, it’s piloting programs that accept data from human-cell assays, organ-on-a-chip systems, and advanced computational models.
This creates a once-in-a-generation opportunity to redefine how we evaluate safety: making development faster, more ethical, and more predictive of real human outcomes.
That’s what the PREDICTS consortium is all about.
What happens next
The project unfolds in two phases: Phase I (beginning this year)
- Building data discovery methods for predictive drug safety models, ingesting data from publications, patents and drug product labels.
- Conducting high-throughput laboratory experiments to interrogate drug permeability, protein interactions, drug metabolism, impacts on gene expression and signaling pathways.
- Compiling data into a composite human model to evaluate the toxicity of a potential drug.
Deep Origin’s technology will power much of this work:
- ADMET-NOW, our suite of 70+ ML models for absorption, distribution, metabolism, excretion, and toxicity.
- Togo, our foundational model for molecular property prediction.
- Balto AI, our conversational assistant for molecular design and simulation.
Phase II will involve testing these models with real drug developers, engaging with FDA regulators for feedback, and refining the models for eventual commercialization.
A moonshot worth taking
Success in moonshots like this is never guaranteed. We may encounter new biological phenomena that challenge our models, or find that the technology isn’t yet ready to fully simulate a human body. But if anyone can do it, it’s this team.
We’re not developing drugs; we’re developing the models that make drug development faster, safer, and smarter. And we’ve already built early versions of many ADMET models, several of which are accessible today through Balto.
The goal is ambitious: a comprehensive virtual human model that can one day reduce or replace animal testing altogether. But we believe it’s achievable - through collaboration across industry, academia, and regulators.
Join us
If you’re a funder, we welcome partnership and discussion as we expand this work beyond ARPA-H’s initial scope. ARPA-H CATALYST funds our in silico safety models for multiple key organs, each of which can become its own product, but our mission is larger than this. We aim to build representations of each key organ and tissue in the body: an actual virtual human capable of capturing the variation of the human population in drug safety.
If you work on ex vivo models that capture human biology in vivo and you’d like to work with us, reach out to us here. While we have selected ex vivo models for our initial proposal, there is a ton of complex human biology to capture as we build a comprehensive virtual human model.
If you’re a drug developer, you can already explore early versions of our models in Balto or reach out for early access.
Together, we can shape a future where drug development is faster, more successful, and no longer dependent on animal testing. If you want to join the cause, reach out!
References
- Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics. Apr 1 2019;20(2):273-286. doi:10.1093/biostatistics/kxx069
- Sertkaya A, Beleche T, Jessup A, Sommers BD. Costs of Drug Development and Research and Development Intensity in the US, 2000-2018. JAMA Netw Open. Jun 3 2024;7(6):e2415445. doi:10.1001/jamanetworkopen.2024.15445
- Kim E, Yang J, Park S, Shin K. Factors Affecting Success of New Drug Clinical Trials. Ther Innov Regul Sci. Jul 2023;57(4):737-750. doi:10.1007/s43441-023-00509-1
- Van Norman GA. Limitations of Animal Studies for Predicting Toxicity in Clinical Trials: Is it Time to Rethink Our Current Approach? JACC: Basic to Translational Science. 2019/11/25/ 2019;4(7):845. doi:10.1016/j.jacbts.2019.10.008
- Schaefer, G. et al. Driving adoption of new technologies in biopharmaceutical manufacturing. Biotechnol. Bioeng. 120, 2765–2770, doi:10.1002/bit.28395 (2023).
- Scannell, J. W., Blanckley, A., Boldon, H. & Warrington, B. Diagnosing the decline in pharmaceutical R&D efficiency. Nat. Rev. Drug Discovery 11, 191–200, doi:10.1038/nrd3681 (2012).
- Booth, B. Culture as a Culprit of the Pharma R&D Crisis, in LifeSciVC (2012).
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
