Owkin announced an agreement on September 11, 2026, to license K Pro, its AI scientist for biopharma research, together with multimodal patient data to Servier, an independent international pharmaceutical group governed by a foundation, to advance Servier’s oncology research.
What the License Covers
Under the agreement, Owkin will make multimodal patient data from its MOSAIC network accessible for analysis to Servier’s research teams within K Pro. The announcement, datelined New York and Paris, framed the license as the next chapter in the companies’ collaboration; their earlier joint work was in patient subgroup identification, also in oncology.
Owkin, which describes itself as an agentic AI company, said K Pro combines multimodal patient data with specialized biological AI agents to support decision-making at each step of the pharmaceutical value chain. According to the company, the platform delivers molecule optimization, target discovery, biomarker identification, subgroup optimization, and clinical trial design, and is trained and validated on real-world patient data and through Owkin’s wet laboratory. Owkin said the system draws on specialized biological tools and a decade of knowledge gained through pharmaceutical partnerships and AI research.
Thomas Clozel, Owkin’s co-founder and CEO, said: “Our AI scientist is changing the way biopharma works. By pairing K Pro with Owkin’s multimodal patient data, Servier’s scientists can investigate the biology of hard-to-treat cancers and reach answers that manual analysis alone cannot deliver.”
Claude Bertrand, Servier’s executive vice president of research and development, said access to high-quality patient data and AI-driven analysis in one place, through K Pro, helps his teams generate and test hypotheses with greater speed and confidence. He said Servier looks forward to applying the capabilities across its oncology research.
K Pro’s Documented Capabilities
Owkin’s product documentation describes K Pro as an AI agent that answers natural-language questions with analysis of multimodal patient data and returns what the company calls rigorous, publication-ready answers. According to the documentation, the platform ranks therapeutic targets and gene candidates using statistical evidence from TCGA and MOSAIC datasets, evaluates target tractability, safety profiles, and therapeutic potential using curated databases, and identifies distinct patient populations based on multi-omics profiles and clinical outcomes.
The documentation says K Pro tests biomarkers and hypotheses across bulk RNA-seq, single-cell, spatial, and clinical data, and that it integrates genomic, transcriptomic, and immune data to identify therapeutic opportunities and predict treatment response. It can also produce statistical reports covering gene and target assessment, biomarker prioritization, and patent landscape analysis.
Owkin says a single intelligent orchestrator sits at the center of the platform, dynamically selecting and combining specialized AI skills so that users work with one agent rather than juggling multiple tools. The company says the agent continuously learns from real-world data, user feedback, and clinical validation.
K Pro is offered in two tiers, per the documentation: a free tier for exploration and demonstration on 19 public datasets, and a professional tier that adds Owkin’s proprietary datasets and the MOSAIC spatial and multiomics dataset, along with data upload, higher usage limits, and deployment either as SaaS or on a client’s own premises.
The MOSAIC Dataset
The patient data covered by the license come from MOSAIC, an Owkin initiative that describes its dataset as the world’s largest spatial multiomics dataset in oncology. The initiative says that by mapping interactions between cells within the tumor microenvironment, it aims to uncover new disease biology, identify more precise patient subtypes, and discover biomarkers and drug targets that could enable more personalized medicine.
The MOSAIC site reports 2,725 total patients across 11 therapy areas, six data modalities, and 100 clinical variables per patient. It reports spatial omics data generated for 2,371 patients and single-cell omics data for 2,168. Its per-modality counts include 2,666 digitized H&E samples, 2,266 whole exome sequencing samples, 2,114 bulk RNA-seq samples, 2,371 spatial transcriptomics samples, and 2,168 single-cell RNA-seq samples, in addition to clinical data for every patient.
By indication, the site reports 532 patients in non-small cell lung cancer, 472 in ovarian cancer, 386 in breast cancer, 315 in glioblastoma, 269 in bladder cancer, 192 in diffuse large B cell lymphoma, 177 in pancreatic cancer, 159 in HNSCC, 100 in colorectal cancer, 96 in mesothelioma, and 23 in gastric cancer. MOSAIC’s founding partners are Owkin, Gustave Roussy, Charité, CHUV, Universitätsklinikum Erlangen, and the University of Pittsburgh.
The October 2023 Partnership
The license extends a relationship the companies formalized on October 17, 2023, when they announced a partnership to apply AI to better-targeted therapies across multiple disease areas, including oncology. The companies said at the time that they would initially take on two challenges in translational medicine and digital pathology, with Owkin applying machine learning to Servier’s clinical data to identify the patient populations most likely to benefit from Servier’s new therapies.
According to that announcement, the first project aimed to identify tumor types, tumor microenvironments, and de-identified patient subgroups that might optimally respond to a Servier asset, with the companies then analyzing tumor evolution to identify drug combinations that could act on additional immune checkpoints or tumor cell-intrinsic mechanisms. The second project explored digital pathology as a way to accelerate screening and enable broader analysis of tissue-based biomarkers. A multidisciplinary team of scientists, computational medicine experts, and clinicians from both companies ran the collaboration between Paris and Boston, and Bertrand held the roles of executive vice president of research and development and chief scientific officer at the time.
Credit: Source link
