
Atlas Bio, the Swiss startup founded by former Google engineers, has raised $12.5 million in a seed round led by Creandum to simulate drug trials using AI foundation models.
Atlas Bio has raised $12.5 million in a funding round led by Creandum to simulate drug trials using AI. Atlas Bio is in Phase II trials, where approximately 70 percent of drug candidates fail.
The Swiss startup plans to grow its scientific and commercial teams, improve its platform, and take on more enterprise customers through 2026. Simon Romanski, one of the first engineers to work on autonomous vehicles in 2014, founded Atlas Bio in 2024 to address the high failure rate in Phase II clinical trials, where around 70 percent of drug candidates do not make it through.
The company has raised $12.5 million so far, including a $10 million seed round led by Creandum. Before the seed round, Atlas Bio raised a $2.5 million pre-seed round, led by First Momentum Ventures and LEA Partners.
The company is headquartered in Basel, incorporated in Delaware, and has a Swiss subsidiary. Atlas Bio plans to grow its scientific and commercial teams, improve its platform, and take on more enterprise customers through 2026. From self-driving cars to drug trials Romanski did his electrical engineering at the Karlsruhe Institute of Technology, EPFL in Lausanne, and Stanford.
His interest in drug development was sparked by living with Type 1 diabetes. Living with diabetes has always made me interested in drug development. It is one of the few diseases where treatment options are improving rapidly.
That progress stands out in a field where most clinical trials fail. In the past, biology and engineering seemed like very distant fields. But I realised that the timing is right for drug development to become more of an engineering discipline, as we are already seeing in molecular design.
It was less a decision to leave self-driving cars and more a decision to finally apply my skills to drug development. Sebastian Bayer, co-founder & CTO, worked on perception models for autonomous driving in a very data-limited environment. He then led Google's research on language models for Search, YouTube, and Google Play.
Dominik Hartl, its chief scientific officer, was a doctoral fellow at the Friedrich Miescher Institute for Biomedical Research and has over 10 years of experience in drug development. Reflecting on how techniques translate across domains, Romanski noted: At the AI model level, very little transfers directly from autonomous driving to drug development. At a more abstract level, though, many of the problems are similar.
Selecting training data carefully is critical. Validation at scale is expensive and comes with legal hurdles.
The winners in both fields have a good strategy for both. I think simulation before deployment is fundamentally the same idea, but it is not a particularly helpful analogy. A more useful parallel is how autonomous driving evolved.
The field started with non-AI methods because people largely underestimated what AI models could do. Over the past decade, almost the entire stack has become AI-driven. I think we will see the same trend in drug development and in our ability to model human biology with AI.
Most biological effects are not caused by drugs. To predict unknown drug effects, we need to learn biology in an unbiased way. By training our foundation model on roughly 2 million human RNA readouts, we can generalise across indications and treatments, enabling simulations that are not tied to one disease or one mechanism.