Owkin unveils biology-focused AI agents for clinical research and drug discovery – News

Biopharmaceutical artificial intelligence startup Owkin Inc. said today it’s making a series of autonomous agents focused on drug discovery and research available to the broader healthcare industry, after cementing partnerships with industry leaders Nvidia Corp. and Anthropic PBC.

The 10-year-old AI company is also releasing its complete agentic infrastructure platform to customers, including a highly curated multimodal patient dataset collected from more than 800 hospitals globally. Whereas other AI firms in the biopharma industry are forced to work with data surfaced through lab research, Owkin says its AI agents are trained on clinically relevant insights from a diverse group of real human patients that reflect real diseases far more accurately.

Owkin says it envisions drug companies and others in the healthcare industry using its AI agents to identify biomarkers, interpret complex medical datasets and guide decision making in clinical trials. Because the agents are built on real patient data, it believes it can improve clinical success rates and accelerate drug discovery, enabling new medicines to reach those who need them faster than ever before, without infringing on anyone’s privacy.

Owkin co-founder and Chief Executive Thomas Clozel said the launch of the company’s agentic infrastructure is a major milestone for the healthcare industry, paving the way for more researchers to access advanced AI tools to aid in drug development. “For the first time, healthcare companies can use specialized agents to quickly and efficiently assist in research, analysis and clinical trial decision making, from their platform of choice,” he said.

According to Clozel, Owkin sees itself as a kind of central connector that links real-world patient data from hospitals, labs and academic institutions via its intelligent AI agents to a pharmaceutical industry that desperately needs more of it. Ultimately, Owkin sees its agentic infrastructure as the foundation of what Clozel calls “biological artificial super intelligence,” or essentially an AI that’s capable of reasoning across the full complexity of biological systems. By accumulating more knowledge over time, Clozel hopes, Owkin will be able to fully automate scientific research and drug discovery to create entirely new treatments and diagnostics.

Owkin’s first AI agent is called Pathology Explorer, and it’s being made available through Anthropic’s Claude for Healthcare Life Sciences offering, partly to showcase the interoperable nature of its agentic infrastructure. Pathology Explorer was built atop of Owkin’s Histo Interpretability Prediction Engine model and is designed to identify different cell types, cell locations and biomarkers from only digital pathology images.

According to Clozell, this is a critical capability that has the potential to accelerate drug discovery and the design of clinical trials. He said it was trained on Owkin’s comprehensive patient dataset to perform high accuracy analysis much faster than any existing model. For instance, it shows 23.7% better classification accuracy on key benchmarks, despite using five times fewer parameters than other agents trained for this type of research. Because the underlying model is much smaller, Pathology Explorer can reduce computational times from weeks to just hours compared to using agents based on larger models.

Owkin’s AI agents will be made available through an application programming interface service and are compatible with the Model Context Protocol, which means researchers will be able to embed them directly into existing healthcare workflows with minimal disruption. The agents will improve over time as they’re continuously refined using new patient data and lab-in-the-loop testing.

The partnership with Nvidia is focused on improving Owkin’s foundational model OwkinZero. The company plans to refine its biological large reasoning model using tools and services such as Nvidia’s Nemo framework and cuCIM, an open-source software library for accelerated multidimensional image processing. Clozel said the goal is to enhance the performance and scalability of OwkinZero’s biological reasoning capabilities and improve its overall robustness.

“By combining Nvidia’s world-class AI ecosystem with our domain specific biological reasoning capabilities and multimodal patient data, we’ll be able to create AI agents that can truly understand and reason about the complexity of biology at a scale that was previously impossible,” Clozell said.

Image: Owkin