Bioptimus announced STELA, the world's largest clinically linked spatial biology atlas, aiming to profile up to 100,000 patient specimens across three continents to power M-Optimus, a foundation model for biology that could reshape how researchers understand disease and develop new treatments. The initiative represents a roughly 20-fold increase in scale over existing spatial biology datasets and marks a significant step toward applying artificial intelligence to biological research at an unprecedented scope. Why Does Biology Need Its Own AI Foundation Model? Foundation models have revolutionized language and image processing by learning from massive datasets. Biology, however, has faced a critical bottleneck: the lack of standardized, high-quality clinical data at scale. While language models like GPT (Generative Pre-trained Transformer) train on billions of text samples, biologists have struggled to assemble comparable datasets that combine molecular, cellular, and clinical information in a unified format. STELA aims to close this gap by generating harmonized datasets that integrate spatial transcriptomics (which maps where genes are expressed in tissues), histopathology imaging, multi-omics data (genomics, transcriptomics, and proteomics), and longitudinal clinical records from real patients. This multimodal approach allows AI models to learn how molecular and cellular interactions drive disease across oncology, immunology, and other therapeutic areas. "Today, most patients' diagnostic data is used to inform decisions for only that individual. We envision a world where every patient can contribute insights to better inform the care and treatment outcomes of future patients," said Jean Philippe Vert, Co-Founder and CEO of Bioptimus. Jean Philippe Vert, Co-Founder and CEO of Bioptimus How Will STELA Generate and Standardize This Massive Dataset? STELA operates through a three-part infrastructure designed to ensure reproducibility and data quality across global research institutions: - 10x Genomics Partnership: The Xenium spatial transcriptomics platform serves as the foundational technology for generating standardized spatial datasets across participating hospitals and research institutions worldwide, ensuring consistency in how data is collected and processed. - Broad Clinical Labs Collaboration: A multi-year agreement leverages Broad's high-throughput laboratory workflows to process biological samples at scale, with co-development of AI-driven quality control metrics and predictive tools to optimize assay performance and automate biological insights. - Global Sample Collection: The initiative will profile specimens across three continents (the United States, Europe, and Asia), with participating institutions contributing samples under standardized protocols and receiving access to rich spatial characterization and foundation model capabilities in return. This collaborative model addresses a fundamental challenge in biomedical AI: ensuring that data generated in one lab is compatible with data from another, and that quality standards don't vary across institutions. By aligning data generation protocols, processing, storage, and AI model development within a unified framework, STELA establishes what researchers call "foundational infrastructure for the next era of biological AI". "Many of the most important questions in medicine come down to understanding how cells interact within complex human tissues. By enabling spatial profiling at unprecedented scale, STELA will generate foundational datasets that allow researchers to connect the underlying biology with disease outcomes," stated Serge Saxonov, Chief Executive Officer and Co-founder of 10x Genomics. Serge Saxonov, Chief Executive Officer and Co-founder of 10x Genomics What Could M-Optimus Actually Do for Drug Development? M-Optimus, the world model of biology that STELA will power, is designed to map how molecular and cellular interactions drive disease. In practical terms, this means researchers could use the model to anticipate how patients will respond to novel therapies, accelerate drug development timelines, and design more effective immunotherapies. The model works by learning patterns from the massive multimodal dataset. If a researcher wants to understand why a particular cancer patient responds differently to a drug than another patient, M-Optimus could analyze the spatial organization of tumor cells, immune cells, and molecular markers to identify the underlying biological differences. This kind of precision medicine approach could reduce failed clinical trials and help match patients to therapies more effectively. Bioptimus is not starting from scratch. The company has already developed H-Optimus, a foundation model of human histology (tissue structure) that has been downloaded over 1 million times and is being adopted across research, drug discovery, and clinical pipelines at 16 of the top 20 pharmaceutical companies. STELA represents the next evolutionary step, integrating spatial and clinical data at a scale that was previously impossible. How Does This Compare to Other AI Breakthroughs in Drug Discovery? The AI-for-drug-discovery space is heating up. XtalPi Holdings, another AI biotech company, reported full-year 2025 revenues of RMB 802.6 million (roughly $110 million USD), a 201% year-over-year increase, and achieved profitability for the first time as a listed AI-for-science company. The company has developed over 200 industry-specific AI models and deployed multi-agent systems that independently orchestrate tens of thousands of compound synthesis experiments per week. XtalPi's approach focuses on automating the chemistry side of drug discovery through robotics and AI agents that can design, synthesize, and test molecules autonomously. STELA, by contrast, focuses on understanding human biology at scale to inform which targets are worth pursuing and how patients will respond. These are complementary approaches: understanding disease biology (STELA's focus) and automating chemistry (XtalPi's focus) together could significantly compress drug development timelines. XtalPi has also expanded into novel drug modalities beyond small molecules, including molecular glues, peptides, and oligonucleotides, and has secured multiple landmark collaborations with cumulative contract values reaching tens of billions of RMB. The company's success underscores growing confidence in AI-driven drug discovery across the pharmaceutical industry. What Are the Real-World Implications for Patients? The ultimate goal of STELA and M-Optimus is to enable precision medicine at scale. Instead of treating all patients with a given disease the same way, clinicians could use AI-generated insights to tailor diagnostics and therapies to individual patient biology. This could mean faster diagnosis, fewer side effects from mismatched treatments, and better overall outcomes. However, realizing this vision requires solving several challenges. Data privacy and patient consent are critical; STELA participants must agree to have their tissue samples and clinical records used for AI training. Data quality must remain high across thousands of samples and multiple institutions. And the AI models must be interpretable enough that clinicians can understand why the model recommends a particular treatment, not just that it does. Bioptimus and its partners are aware of these challenges. The partnership with Broad Clinical Labs specifically includes co-development of quality control metrics and predictive tools designed to ensure data integrity. By combining industrial-scale data generation with frontier AI research, the collaboration aims to build STELA on a foundation of "unprecedented technical precision," according to Niall Lennon, Chief Scientific Officer of Broad Clinical Labs. "To unlock the true clinical potential of spatial biology, we must pair massive-scale data generation with uncompromising data quality. By combining our high-throughput laboratory workflows with Bioptimus's advanced AI, we are co-developing next-generation quality control metrics that ensure the highest data integrity," noted Niall Lennon, Chief Scientific Officer of Broad Clinical Labs. Niall Lennon, Chief Scientific Officer of Broad Clinical Labs STELA's launch signals a broader shift in biomedical AI: from building models on small, curated datasets to constructing foundation models trained on massive, real-world clinical data. If successful, this approach could accelerate the pace of drug discovery and usher in a new era of AI-informed precision medicine.
