The Basel startup says its AI model of human biology can predict how patients will respond to drugs before a live trial begins. The money behind it is easier to count than the proof.
Atlas Bio, a Basel-based startup that simulates clinical trials with artificial intelligence, came out of stealth on Oct. 27, 2026, with a seed round led by the Swedish venture firm Creandum. The size of that round depends on who is reporting it: Dealroom and Tech Funding News list $10 million, while Creandum's own announcement says $11 million.
No public clarification from the company appears in the materials reviewed, and the discrepancies do not stop there. Tech Funding News and Reuters put total funding at $12.5 million, a sum that includes a previously undisclosed $2.5 million pre-seed round led by First Momentum Ventures and LEA Partners. Chief executive Simon Romanski, writing on LinkedIn, cited $13.2 million in "combined funding" from Creandum, First Momentum Ventures and what he called LEA Ventures, a slightly different name for the pre-seed investor. The materials reviewed do not explain where the extra several hundred thousand dollars comes from.
What Atlas Bio sells
Atlas says it runs in silico clinical trials in immunology, meaning the experiments happen in software rather than in patients. A foundation model of human biology, the same broad category of technology that underpins large language models but trained on biological data, predicts and compares how patients respond to different treatments and different diseases. Pharmaceutical teams would use the output to inform decisions on clinical development, licensing and a drug's lifecycle before they commit to a live study.
Creandum's post said the model was trained on roughly 2 million RNA readouts from human tissue. Reuters reported that a simulation takes 12 weeks, and that results from a first trial are expected within six weeks. The two timelines may describe different things, though the reporting reviewed does not say so.
The pitch rests partly on a point about design. Atlas said its simulations can compare several drug mechanisms "in the same patients at the same disease state," something no live trial can do, since a patient cannot be enrolled in two arms at once. The company also says it returns a full transcriptomic, patient-level response (a readout of gene activity, patient by patient) rather than a single score.
Case studies, with caveats
The company's website describes simulated cohorts in asthma, atopic dermatitis, ulcerative colitis and alopecia areata. In atopic dermatitis, Atlas said it predicted that an MRGPRX2 inhibitor would miss its endpoint, and that a later Phase II readout confirmed the miss. The customer was not named. Atlas said detailed methods and readouts are available only under a nondisclosure agreement, which means outsiders cannot check the work.
In ulcerative colitis, the company said it simulated a PD-1 depleting antibody across 163 patient profiles and predicted remission of up to 53.4%.
Creandum's post went further. It said Atlas is under contract with a top-10 global pharmaceutical company and in talks with more than a dozen of the largest drugmakers, and that "every prospective prediction made on real trial outcomes has proven correct." Those are claims from an investor with a stake in the outcome. No customer has confirmed them publicly.
Valuation and backers
Handelsblatt reported on Oct. 28, 2026, that the company is valued at $60 million, and said angel investors include the founders of Black Forest Labs and Chai Discovery. An aggregator cited a post-money valuation of roughly $100 million, attributing it to TechCrunch, but that report could not be located. Neither figure has been confirmed by the company in the materials reviewed.
Katharina Wilhelm, a Creandum partner, wrote: "We're proud to be leading Atlas Bio's $11m Seed round." AI4S.fyi and Sesamers said she joins the board. Dealroom ranked the seed in the 98th percentile for Swiss big data seed rounds.
The team and what comes next
Romanski previously led machine learning teams at understand.ai. Chief technology officer Sebastian Bayer led large language model research at Google, according to the company's website. Dominik Hartl, the chief scientific officer, built an indication-selection framework at Idorsia, and chief commercial officer Malcolm Fogarty previously worked at Lindus Health, GSK and Haleon.
Atlas was founded in 2024 and registered as Atlas Bio AG in Basel on July 30 of that year. As of Oct. 28, 2026, LinkedIn lists 11 to 50 employees. According to Sesamers, the new money will fund hiring of scientific and commercial staff and deeper validation of the platform, and will help bring on more enterprise customers through 2026.
A crowded field, and a hard problem
Atlas is not alone. Unlearn, which builds digital twins for clinical trials, raised a $50 million Series C in February 2024, according to Dealroom. QuantHealth, another trial-simulation company, announced a $45 million Series B led by Qumra Capital. Both figures are drawn from the reporting reviewed, and readers should check for later rounds.
Atlas is aiming at Phase II, the stage where drug candidates fail most often. A BIO analysis covering 2006 to 2015, now an older benchmark, found that 30.7% of drugs moved from Phase II to Phase III. Regulators are also showing interest in alternatives to traditional testing: the FDA has issued draft guidance on new approach methodologies, a category that includes in silico modeling.
Romanski told Handelsblatt that the decisive question is no longer whether good molecules can be found, but how quickly their efficacy can be proven in clinical trials. Atlas said its first trial results are due within six weeks of the announcement, which will offer the first public test of that argument.
