Axiom Winter 2025 Investor Memo

Axiom: Enable scientists with the most accurate models for human drug toxicity and exposure.

Axiom is enabling scientists to eliminate drug toxicity by replacing traditional experimentation with AI models. Our predictive safety models are more accurate, affordable, and interpretable than traditional experiments like animal studies. By accurately predicting human safety risks early in discovery, we aim to drastically reduce the 90% clinical failure rate.

We are starting with drug-induced liver injury which causes 25% of clinical trial failures, 30% of post market withdrawals, and 55% of recent FDA warnings. We are positioned as the leading AI/ML partner amidst strong FDA regulatory tailwinds to replace animal testing.

• Funding: Series A
• Investors: Amplify Partners, Dimension Capital, CRV, Abstract VC, Zetta VP, Jeff Dean.
• Traction: 5 paid pilots with large pharmaceuticals.

My cofounder and I tried to use fancy ML/AI methods to accelerate drug discovery but we failed. Every time we would progress a drug towards clinical trials, many would fail in the late stages because of toxicity. Turns out, this is an industry wide problem with many drugs failing due to toxicity. This rate of failure is costing the industry tens of billions of dollars every year.

The crazy thing, it's not just the failures. Approved drugs can also be very toxic. Have you ever known someone with cancer? Cancer drugs have massive side effects which make patients think twice about taking them. Accutane, the best acne medicine, causes significant adverse events. It's not just the drugs, we are surrounded by potentially toxic chemicals. 

Toxicity is the primary reason we lack an abundance of cheap, effective drugs for conditions like aging, obesity, chronic pain, and depression. For many medicines, the risk of toxic effects outweigh the benefit in these daily conditions. Assessing this risk-benefit equation in clinical trials is what makes drug discovery so expensive and prone to failure.

Existing animal models for testing toxicity are not great. They are inaccurate, slow, expensive, and ethically dubious. Scientists lack the tools to predict toxicity effectively, forcing a cycles of repeated failures. We are still using the same toxicity testing methods invented in the 60s - rats, dogs, and monkeys.

Axiom is changing this paradigm. We build AI models that predict toxicity with greater accuracy than traditional experimentation, replacing or augmenting these outdated methods. We want our models to be more accurate and interpretable than any other solution in the world. We want to develop this capability for all major molecular modalities (small molecules, antibodies, peptides) and across all major tissues (liver, heart, kidney, GI, brain, etc.).

Within three years, we want to be the first company to enable the replacement of animal testing in a major novel investigational new drug application. As our technology is adopted over the next 5-10 years, we want the risk of a new drug causing toxicity to approach zero. This will virtually eliminate clinical trial failures due to toxicity, dramatically reducing the overall failure rate of new drugs and changing the entire economics of drug discovery.

By eliminating the risk of toxicity in new drugs, we'll unlock the opportunity for society to completely rethink drug development and validation. For the first time, we will have an overabundance of safe drugs, agrochemicals, cosmetics, food additives, and more.

We started Axiom by recognizing both an early societal desire to phase out animal testing and the technological trends poised to make it possible. Just 18 months later, these societal pressures have dramatically intensified. We're now seeing clear signs that this momentum will exponentially accelerate, with major institutions across the pharmaceutical, cosmetic, agrochemical, industrial, and regulatory sectors moving in unprecedented unison to change over 50 years of established precedent.

The shift away from animal testing in drug development gained significant momentum in 2023 when Congress dropped the legal requirement for new drugs to undergo animal testing. This was replaced with the option to use “cell-based assays, organ chips, and computer modeling”. In April 2025, the FDA announced its intention to phase out animal testing over the coming 3-5 years, with the HHS Secretary telling the president it's one of his top priorities. Since then, the NIH, FDA, and HHS have only doubled down, removing funding for grants that rely on animal testing, canceling NIH animal experimentation, and replacing key animal research leaders with proponents of next-generation, non-animal approaches. Congress is now preparing to further support this transition with the introduction of the FDA Modernization Act 3.0 which we expect to pass in 12 months. When Axiom visited Congressional members, we were told that replacing animal testing is "the only bipartisan issue in Congress" and widely seen as a "must-have easy win."

Recent regulatory shifts are beginning to significantly impact the industry, creating momentum for alternatives to animal testing. Axiom is actively involved in major public-private initiatives which bring together the FDA, NIH, big pharma, cosmetics, and agrochemical companies, along with non-profits and CROs, to evaluate and advance new approaches. These efforts are being driven with a level of urgency comparable to the COVID-19 era. Axiom is uniquely positioned as the only company in these initiatives capable of applying machine learning at scale to this problem. Beyond just training the models, we are also leading when it comes to defining the benchmarks and evaluations for evaluating ML/AI on modern chemistry.

Axiom's AI/ML leadership in toxicity prediction is attracting significant attention. Major pharmaceutical companies are increasingly inviting us to present our data at key conferences and internal team presentations, discussions that are now converting into large paid pilots. As enterprises try to keep up with the FDA’s changes and advanced their own initiatives, they see Axiom as doing some of the most promising work in the space. We're actively collaborating with top pharma, cosmetics, agrochemical firms, and contract research organizations (current largest providers of animal testing). Axiom isn't here to make wild claims or pretend to have all the answers. Our role is to provide nuanced truths, empathizing with drug hunters about the challenges of this shift while inspiring them to partner with us for a far better future. These paid pilots are crucial. They're about working directly with large industry institutions to honestly assess where we stand today and chart a clear path for the scientific progress we need to see in the future. Its a collaboration betweeen Axiom and pharma at every level from benchmarking, choosing compounds, design experiments, reviewing analysis, and making final conclusions. Companies see Axiom as the only partner genuinely approaching ML/AI in toxicity the right way. We will continue working closely with them until they are confident that using Axiom will dramatically improve the clinical success rates of their drugs.

The essence of getting drugs approved involves demonstrating to the FDA that the benefits outweigh the risks of toxicity. A significant factor contributing to the 90% drug failure rate is unacceptable toxicity which causes many drugs to fail when risks surpass potential benefits. A key reason for this high failure rate is that drug developers today lack the ability to accurately model and understand toxicity during drug development. Current toxicity testing methods have low predictive accuracy for human outcomes, are extremely expensive and slow, difficult to interpret, and challenging to use for iteratively improving drug candidates.

This has resulted in an ineffective drug development paradigm where toxicity and safety evaluations are rarely incorporated into the iterative and developmental phases of drug discovery. Typically, comprehensive drug safety testing only occurs at later stages, relying heavily on costly, time-consuming, and imprecise animal models. This approach explains why 59% of drugs fail late-stage preclinical toxicity testing after over $10 million has been invested. It is also the primary reason 40% of drugs fail during clinical trials due to toxicity after over $20 million has been invested. Consequently, the pharmaceutical industry loses tens of billions of dollars every year from toxicity-related failures. Pfizer experienced the failure of two oral obesity drugs due to liver toxicity, while Eli Lilly successfully advanced a safe oral obesity drug, resulting in Eli Lilly gaining over $100 billion in market capitalization. However, Eli Lilly has also experienced liver toxicity in some of their historical drug candidates. Similarly, in a single week in 2024, over $5 billion in market capitalization was lost as several clinical trials failed due to liver toxicity. As of mid 2025, these companies' stocks have yet to recover. Additionally, three biotechnology firms recently declared bankruptcy after their drug candidates failed clinical trials due to liver toxicity. And in 2025, it was reported that a company's product was associated with two to three deaths due to acute liver injury leading to a strong FDA response.

This is why Axiom is partnering with drug hunters to establish a new paradigm where drug toxicity can be understood and addressed in the earliest stages of development, before it's too late.

Axiom is building three core foundation models to enable this new paradigm:

1. Predict human drug exposure and metabolism: Accurately model how a compound is absorbed, distributed, metabolized, and exposed across all major human organs.
2. Predict human organ-level safety and toxicity: Assess whether those exposures are safe or toxic for each major organ, starting with the liver.
3. Reason about FDA risk–benefit assessments: Contextualize predicted safety and efficacy profiles to estimate how regulatory bodies, like the FDA, would evaluate a compound’s overall risk–benefit balance.

At the earliest stages of drug discovery, Axiom’s models will take in experimental data which quantifies a drug’s entire life cycle from exposure to metabolism to cell/organ-level function. It will use this data to accurately predict a drug’s human toxicity/exposure profile, reason through the mechanism, then suggest next steps for scientists to take. In the later stages with a clear target disease indication and patient population, Axiom’s reasoning agent will then use the predicted profile to reason about how the FDA might conduct its risk-benefit assessment.

This foundational work represents Axiom’s first phase: creating a toolkit that accurately predicts human drug toxicity and exposure. Each model involved is highly complex and must integrate multiple data modalities to generate accurate outputs. For example, predicting liver toxicity requires data on a drug’s exposure (tier 1 ADME), metabolism (reactive metabolite assay), and organ biology (2D+ Multi-cell Hepatic System). We may generate this data experimentally then run it through models or predict the data computationally from the chemical structure alone. The computational prediction requires modeling chemical structures using graph neural networks, liver biology using vision transformers, and clinical trial data using large language models.

Once Axiom has developed a highly accurate toolkit for modeling human outcomes, reasoning agents can leverage it to approximate the mechanism, compare it to well known references, and reason through a risk-benefit assessment. To do this effectively, Axiom’s agent must excel at tool use of all kinds, especially in applying Axiom’s models to predict human outcomes and understanding the larger context of biology, chemistry, and clinical development. It must determine which models to invoke, accurately interpret the predictions, and contextualize those predictions with respect to uncertainty, probability, and domain relevance. The agent will also have access to Axiom's reference database which contains thousands of molecules fully annotated with their clinical outcomes and profiled across all of Axiom's experiments. How the model effectively uses these references to provide the scientists with the most impactful information is an active area of research.

With these predictions in hand, the agent can then reason from first principles to inform drug hunters on the biological and chemical mechanisms driving the predicted outcomes, what next steps to take to derisk their drug, and how the FDA will reason about the risks/benefits. The agent will be trained to assess the mechanism behind the risk from first principles. It will have a foundational understanding of various chemistry, biology, disease indications, disease severity, and the degree of benefit patients experience from improvements in disease state. It will then use this foundation to reason through the sequence of steps which cause toxicity and cause the risk to be higher than the benefit. This benefit must then be weighed against the predicted, or observed, toxicity risk. At a basic level, the agent should recognize that humans tolerate higher-risk drugs for life-threatening conditions like cancer, but have a much lower risk tolerance for chronic conditions such as obesity or insomnia. Of course, this is the simplest case. The real challenge lies in the gray areas, where trade-offs are less clear-cut and the biochemical mechanisms may not be measurable with existing tooling.

To effectively communicate complex scientific insights to drug hunters, text alone is insufficient. The agent must be able to generate and interact with rich, intuitive data visualizations that translate predictions into clear, actionable insights. For instance, visualizing shifts in liver toxicity across escalating exposure levels, surfacing trends across molecular properties, or identifying toxicophores within chemical structures can be essential for informed decision-making. These capabilities demand tight integration between the agent and a powerful visualization layer. Equally critical is the system’s usability. Scientists should experience an interface that feels fast, intuitive, and even delightful to use—reducing cognitive load and enhancing their trust in the platform.

This agent and its frontend must also be deployed through a robust, enterprise-grade SaaS platform which is capable of operating behind enterprise firewalls and meeting strict compliance standards of the world’s largest pharmaceutical companies. Deployment should be seamless, secure, and extensible across scientific teams and workflows. It’s not enough for Axiom to build highly accurate AI and elegant product interfaces, we must also meet rigorous enterprise deployment requirements. Part of our land and expand strategy will involve deploying our enterprise SaaS to one team then expanding the platform into many other teams. As it expands, the platform will integrate with additional scientific data and workflows from other teams across the drug discovery process. This data will then be used to improve the overall enterprise experience through custom, optimized agents with better context.

We will start by building these benchmarks and technology for small molecule liver toxicity then we will expand to additional organs and modalities. https://axi.om/products/liver

Axiom has invested millions of dollars to build massive datasets to solve this problem. We focus on pairing massive experimental datasets which quantify a drug’s exposure, metabolism and organ biology with internet-scale datasets of human adverse event outcomes. We developed proprietary in-house wet lab protocols to run "flattened spheroid" experiments profiling over 100,000 drugs. In parallel, our LLM agents have curated and reasoned over global human clinical trial data on liver toxicity from more than 12,000 drugs: https://axi.om/dataset. Openly sharing the details of this dataset with industry has been a major contributor to the traction and performance we’ve experienced to date.

Going forward, we have the opportunity to build fully automated labs which serve as the “Tesla Factories” of human toxicity data. However, making the investment in this infrastructure is a massive decision. It's a decision that has a lot of upside but it's also the only thing that can make us go to zero pretty quick due to its capital requirements. So we first need to decide what datasets we want to invest in. We do this by solving the equation:

"Dataset X increases model performance by Y or adds Y new model capabilities and this generates Z value for customers".

This is a really hard equation to solve which requires many different perspectives and expertises. Hypothesizing about a solution to this equation is just the first step as we then need to decide what the best path is to iteratively acquire these datasets. Should we contract it out to another lab, build our own “Tesla Factory” lab, or partner with another company to acquire the data? Having many people with different expertises who are extremely thoughtful about this and hungry to think in this way is important. It's not just about building a lab for the sake of building a lab. That's easy and has been done many times before. Thinking through and executing on the above strategy is what's really hard and has never truly been done. Being truly great at this will be one of Axiom’s biggest strengths and largest technological moats. We will not just build great datasets but we will also deeply understand the relationship between data, model performance, and customer value enabling us to get smarter about capital allocation over time.

Our go-to-market strategy and initial customer relationships follow a stepwise approach. We are starting with small molecule liver toxicity—a major challenge in the industry, responsible for 25% of clinical trial failures and 30% of post-market drug withdrawals. Our goal is to become globally recognized as the leading solution for liver toxicity and the go-to partner for any related issues. We’re already seeing early signs of success. Despite being a very new company, and competing against long-established players, companies are proactively reaching out to us for help with their liver toxicity problems. They often find us through word of mouth and our scientific publications, and consistently tell us that no one else has been able to solve their issues the way we have. All of our initial pilots and relationships are focused on evaluating and testing our liver toxicity models.

• Closed paid enterprise studies with 5+ top 10 pharmaceutical companies

For these pilots, our goal is to validate our performance, value proposition, and long-term vision in order to convert these companies into recurring users. Landing larger contracts isn’t just about where we are today. It's equally about telling a compelling story of where we’re headed, especially in the context of what we currently lack and where we underperform. Large enterprises, particularly in scientific R&D, often want to collaborate on building the future together. Once we land a pilot, there are many paths to expand our relationship with pharma. Eventually, we want to expand to all the major organ systems and modalities which creates value for our customers.

To accomplish its mission, Axiom must execute at the intersection of machine learning, software engineering, chemistry, biology, and product design. We are building the best team in the world at this intersection. We don’t believe in only hiring people with domain expertise. We take inspiration from places like Renaissance Technologies and Genentech who hired outsiders that were extremely talented, hard working, curious, and had deep empathy for learning the challenges in a new space.

We only want to hire people who deeply inspire us and relentlessly push us to be better. They must level up the entire team, bringing significant energy and a sense of awe. They should have great taste for what matters, see what needs doing, and do it with great urgency. They should just act. There must be a deep curiosity for all facets of Axiom, a greater ambition fueling their excitement for what we do. They need to be cracked, technically excellent, obsessive masters of their craft. They're irrational; they could work in big tech, but they know it will not satisfy them: they want the challenge of solving a brutally difficult scientific problem and building a generational company from the ground up, as well as the rewards and glory that will come as a result.

• Total funding: Series A
• Major Investors: Amplify Partners, Dimension Capital, CRV, Abstract VC, Zetta Venture Partners, Jeff Dean
• Target runway: 2+ years with clear path to profitability
• 16 full-time employees
• Growing to 20-22 employees with the Series A
• 2 poison dart frogs
• Product launched April 2025
• 5+ closed enterprise pilots