Swiss Startup Atlas Bio Raises $12.5M to Simulate Drug Trials

Your daily brief on leadership and business

Swiss startup Atlas Bio has raised $12.5 million to build an AI foundation model that runs clinical trials in silico. Creandum led the round, and the company also disclosed a previously undisclosed $2.5 million pre-seed round led by First Momentum Ventures and LEA Partners. Creandum partner Katharina Wilhelm joins the board.

The model is trained on roughly 2 million RNA readouts from human tissue. These are the molecular signals that show how genes switch on and off in the body. From them the model simulates how a drug behaves in each person within a virtual cohort of patients. Atlas Bio says this can replace Phase II trials and the animal studies that come before them.

The failure rate it targets

Phase II trials cost drug companies more than $100 million and take two to three years. Around 70 per cent of drugs fail at this stage. Only 30 per cent show an effect on the disease chosen for the trial, yet among those that are eventually approved, about 60 per cent turn out to treat a different condition better than existing treatments.

Atlas Bio runs trials across all diseases at once. That shows which patients actually benefit before a sponsor commits to a narrow Phase II design. Its in silico trials take the form of a 12-week simulation project. Because the trial is virtual, several treatments can be tested on the same patient profile at the same disease state, which a live trial cannot do.

The company was founded in 2024 by Simon Romanski, its CEO, and Sebastian Bayer, its CTO. Bayer previously led large language model research at Google. Dominik Hartl, a former doctoral fellow of the Friedrich Miescher Institute for Biomedical Research, leads scientific strategy.

Atlas Bio already works with a top 10 global pharma company and is in talks with several others. A Basel-based biotech client has dropped animal testing from one drug development workflow. It went straight to human trials on the basis of Atlas Bio's simulation output.

Romanski said that for sixty years, drug development has rested on two flawed assumptions: that mice predict humans, and that the only way to learn whether a drug works is to give it to thousands of patients and wait. Bayer said that training on human RNA data lets the model generalise across indications and treatments, so simulations are not tied to one disease or mechanism.

The company notes that the FDA Modernization Act 2.0, passed in 2022, removed the legal requirement for animal testing before human trials. The new funding will go to scientific and commercial teams, deeper platform validation and onboarding more enterprise clients through 2026.

Source: Tech.eu