Atlas Bio Raises $12.5M to Replace Phase II Drug Trials With Virtual Patients
Key Takeaways
- Atlas Bio virtual drug trials have raised $12.5 million in funding led by Swedish venture capital firm Creandum, an early backer of Spotify and Trade Republic.
- The Swiss startup aims to replace Phase II drug trials, which it says cost more than $100 million, take two to three years, and see about 70 percent of drugs fail at that stage.
- Its AI foundation model, trained on about two million human RNA readouts, would replace those trials and animal testing with 12-week simulations.
- Virtual trials can also evaluate drugs across multiple diseases simultaneously, something the company says is not feasible in conventional clinical trials.
- Results from a first trial are expected within six weeks, which will be the first real test of the company’s claims.
- Atlas Bio was founded by former Google employees alongside AI and pharmaceutical specialists.
Atlas Bio, a Swiss AI startup, has raised $12.5 million to pursue one of the boldest ideas in drug development: replacing Phase II clinical trials with virtual patients. The funding round, announced on 27 October 2026 and led by Swedish venture capital firm Creandum, backs a company that wants to compress years of human testing into 12-week computer simulations. If it works, it could change how medicines reach patients. If it does not, it will join a long list of technologies that promised to transform drug discovery and delivered less.
The stakes are high because Phase II is where most drugs die. According to Atlas Bio, Phase II trials cost more than $100 million, take two to three years, and see about 70 percent of drugs fail at that stage. Those figures are the company’s own, and they describe the problem it is trying to solve: drug development is slow and expensive largely because the testing stage is slow and expensive. A simulation that could predict, with regulatory-grade reliability, how a drug will perform in humans would remove the largest bottleneck in the pipeline.
What a Virtual Trial Actually Means
A virtual trial is not a video game version of a clinical trial. It is a computational model of human biology, built from real molecular data, that predicts how a drug will behave in the human body. Atlas Bio’s AI foundation model is trained on about two million human RNA readouts. RNA readouts measure which genes are active in a cell at a given moment, and they reveal how cells respond to disease and to drugs. By learning from millions of those readouts, the model builds a picture of human biology detailed enough, in the company’s account, to simulate a drug’s effects without giving it to a person.
The proposed workflow runs like this. Instead of recruiting patients, dosing them, and waiting months or years for outcomes, researchers run a 12-week simulation on virtual patients. The model predicts efficacy and safety signals across different patient profiles. Because the trial is computational, it can also evaluate a drug across multiple diseases simultaneously, something the company says is not feasible in conventional clinical trials, where each disease requires a separate trial with separate patients, protocols, and approvals.
That multi-disease capability, if real, is as important as the speed. Many drugs have effects beyond their original target disease, but testing those effects conventionally is so expensive that most are never explored. A virtual trial that can screen a drug against several conditions at once could uncover new uses for existing compounds, which is one of the cheapest ways to find new treatments. It could also flag safety problems earlier, before a drug enters human testing at all.
It is essential to be precise about what Atlas Bio is claiming. The company aims to replace Phase II trials and animal testing with simulations. That is an aim, not an accomplished fact, and not a regulatory decision. Regulators decide what evidence is sufficient to approve a drug, and no regulator has approved a medicine based solely on a virtual trial. The company’s first trial results, expected within six weeks, will be the first public evidence of whether its simulations predict real outcomes. Until those results appear and are scrutinized, virtual trials remain a promising approach, not a proven replacement.
Why Phase II Is the Right Target
Drug development has three clinical phases before approval, and each has a different job. Phase I tests safety in a small group of people. Phase II tests whether the drug works, and at what dose, in a larger group of patients with the disease. Phase III confirms those results in a much larger population. Phase II is the valley of death because it is the first time a drug faces real patients at scale, and most drugs fail there. A compound that looked promising in animals and in early safety testing often turns out to lack efficacy, or to have side effects that only appear in sick patients.
That failure pattern is why Phase II is so expensive. The cost is not just the trial itself. It is the years spent reaching it, the manufacturing scale-up, the patient recruitment, and the clinical infrastructure. When 70 percent of drugs fail at this stage, as Atlas Bio says, the cost of the failures is carried by the minority that succeed, which is one reason approved medicines are so expensive. A reliable simulation that could predict Phase II outcomes before a human trial begins would save money in two ways: by killing doomed drugs earlier, and by letting promising drugs move faster.
The animal testing angle matters too. Drugs are currently tested in animals before humans, but animal biology differs from human biology in ways that limit prediction. A model trained on human RNA data, rather than animal data, could in principle predict human responses better than an animal model can. That is the scientific bet behind Atlas Bio. Human molecular data, at sufficient scale and quality, may be a better guide to human outcomes than any animal proxy. The two million RNA readouts in the training set are the evidence offered for that bet, though the quality and diversity of those readouts matter as much as their number.
There are obvious objections, and they deserve space. A model trained on RNA readouts captures one layer of biology: gene activity. Human disease also involves proteins, metabolites, cell interactions, organ systems, and environmental factors that RNA data alone does not capture. A simulation that predicts well on average may still miss rare side effects, drug interactions, or effects in patient subgroups that are underrepresented in the training data. Those are exactly the failures that Phase II trials are designed to catch, and a virtual trial that misses them could be worse than no trial at all if it creates false confidence.
Regulators will also ask hard questions. A clinical trial produces auditable evidence: protocols, patient records, adverse event reports, and independent oversight. A simulation produces model outputs, and the audit trail runs through training data, model architecture, and validation methods that are harder to inspect. For virtual trials to be accepted, even partially, the field will need standards for model validation, transparency about training data, and independent replication of results. Atlas Bio’s first trial results, due within six weeks, will be judged against those expectations, not just against the company’s own claims.
The Investor Signal
Creandum led the round, and its track record is part of the story. The Swedish venture capital firm was an early backer of Spotify and Trade Republic, which gives it credibility in picking companies that reshape established industries. That credibility does not guarantee Atlas Bio’s science, but it signals that experienced investors found the team and the approach convincing enough to fund. The company was founded by former Google employees alongside AI and pharmaceutical specialists, which is the combination a project like this needs: machine learning expertise from the technology world and domain expertise from drug development.
The funding amount is also informative. At $12.5 million, this is an early-stage round, not the kind of large financing that would fund a full drug pipeline. It is enough to run the first trials, validate the model, and build the evidence base for a larger round. The company’s timeline reflects that: first trial results within six weeks, then presumably a larger raise if those results are strong. Investors, regulators, and potential pharmaceutical partners will all be watching the same milestone.
For readers who follow AI in biology, this story connects to work we have already covered. Our article on Hugging Face Carbon-X, the model that lets you chat with DNA shows how AI is making genetic data accessible and interpretable. Atlas Bio applies a similar idea, models trained on molecular data, to a different problem: predicting drug outcomes instead of answering questions about DNA. And for the clinical side of health AI, our piece on AI scribe errors in healthcare is a useful reminder that AI in medicine must be judged by its failures as well as its successes. Virtual trials will be judged the same way.
What to Watch Next
The six-week horizon is the key date. Results from the first trial will show whether Atlas Bio’s simulations produce predictions that hold up against real-world comparison, and how the company presents those results will matter as much as the results themselves. Independent scrutiny, peer review, and comparison against conventional trial outcomes would strengthen the claims considerably. A press release with selective highlights would do the opposite.
Beyond that, watch for pharmaceutical partnerships. A virtual trial platform is only useful if drug companies use it, and their adoption will depend on whether the simulations save them time and money without adding regulatory risk. Early partnerships, even small ones, would signal that the industry takes the approach seriously. Regulatory engagement is the other signal: any indication that agencies are willing to accept simulation evidence as part of a submission, even as supporting evidence, would be a meaningful step.
The longest view is the most interesting. If virtual trials work, drug development could become faster, cheaper, and more exploratory. Drugs that are currently abandoned after Phase II failures might be rescued by better prediction. Rare diseases, where patient recruitment is hardest, could become more viable targets. And the cost of approved medicines could fall if the failure rate at Phase II falls with it. Those are large promises, and they depend on a chain of claims that starts with a model trained on two million RNA readouts and ends with a regulator accepting a simulation in place of a trial. Atlas Bio has raised $12.5 million to start proving that chain. The proof will take much longer than six weeks, but the first results will tell us whether the chain is worth following.
