DeepMind Alumni's AI Biotech Lands Third Major Cancer Data Deal in Five Months

Paris-based AI biotech Bioptimus is accelerating its ambitions in cancer diagnostics with a new partnership that brings UK National Health Service patient data into its growing biological foundation model. The company announced a collaboration with Cancer Research Horizons that will integrate profiles from over 1,750 colorectal cancer patients into its M-Optimus AI platform—the third such data agreement in just five months.

Founded by former Google DeepMind and Owkin scientists and launched in 2024 with $35 million in seed funding led by Sofinnova Partners, Bioptimus is building what it calls a comprehensive atlas of cancer through its STELA (Spatial Tissue Embedding Learning Atlas) initiative. The goal: profile 100,000 patient specimens through a network of research institutions to train AI models capable of predicting treatment responses and patient outcomes.

Randomized Trial Data Sets This Partnership Apart

What distinguishes this latest agreement is the nature of the data itself. Bioptimus gains access to the S:CORT (Stratification in Colorectal Cancer Trial) dataset, which Cancer Research UK spent two decades assembling. Unlike observational datasets that simply track what happened to patients, S:CORT contains randomized, protocol-controlled treatment arms—a feature that could prove critical for training AI to distinguish which patients will benefit from specific therapies versus those with poor prognoses regardless of treatment.

The dataset spans the full disease spectrum from precancerous adenomas through stage 2-3 disease to advanced first-line treatment. It includes H&E whole-slide images, genomic information, copy-number variation, microsatellite instability calls, RNA expression, DNA methylation, immunohistochemistry, and clinical outcomes—the kind of multimodal data that biological foundation models require to make clinically meaningful predictions.

According to David Cahané, Bioptimus cofounder and general manager, the partnership represents the company's first UK patient cohort for STELA, adding NHS-treated patients to existing US, European, and Asian data. This geographic diversity could help ensure that AI models generalize across populations and healthcare systems, potentially reducing disparities in diagnostic accuracy.

Building Momentum in a Competitive Field

Bioptimus launched STELA in March 2026 with partnerships involving 10x Genomics and Broad Clinical Laboratories. In June 2026, the company added a collaboration with Fédération Francophone de Cancérologie Digestive that brought profiles from 3,000-plus gastrointestinal cancer patients into the fold. The rapid cadence of data agreements positions Bioptimus as a fast-moving competitor in the biological foundation model space, where it faces rivals including Absci, Boehringer Ingelheim, IBM, and Genesis.

The company's approach centers on molecular inference, prognosis prediction, and treatment-response prediction capabilities that could help oncologists make more informed decisions about colorectal cancer care. Bioptimus plans to test M-Optimus on unseen trials, validate against independent datasets, and allow pharmaceutical partners to evaluate the model using their own protocols—a validation strategy designed to build confidence in the platform's real-world utility.

Implications for Precision Oncology

For the broader field of AI-driven cancer diagnostics, Bioptimus' standardized data ingestion pipeline and shared ontology approach could establish industry standards for how multimodal cancer datasets are assembled and used to train foundation models. The emphasis on randomized trial data rather than purely observational information may also set a new bar for the quality of evidence required to support AI treatment predictions.

As biological foundation models move from research curiosity to clinical tool, the race is on to assemble the most comprehensive, diverse, and high-quality training datasets. With three major partnerships in five months and a clear path to 100,000 patient specimens, Bioptimus is positioning itself at the forefront of that competition.

Sources

1. Bioptimus, Cancer Research Horizons partner on AI cancer model — MobiHealthNews