Two healthcare technology pioneers—Orakl Oncology and Mendel AI—have achieved breakthrough performance improvements using Meta's open-source AI models, with Orakl reporting 26.8% higher prediction accuracy for cancer treatments and Mendel reducing clinical trial matching time from hundreds of days to just one day.
In the world of healthcare innovation, two critical bottlenecks have long frustrated progress: accurately predicting which cancer treatments will work for specific patients and efficiently matching patients to appropriate clinical trials. Two companies leveraging Meta's open-source AI models are dramatically accelerating these processes, potentially saving millions in development costs while bringing treatments to patients faster.
Orakl Oncology, a spinoff from Europe's renowned Gustave Roussy Institute, faced the challenge of analysing vast amounts of imaging data from lab-grown cancer cells called organoids. These organoids simulate how drugs might perform on actual patients, but extracting meaningful predictions from the imagery required quantitative analysis that specialised models couldn't provide efficiently.
The solution came through implementing Meta's open-source DINOv2 computer vision model, which had demonstrated remarkable capabilities in learning from diverse image collections. Rather than building a specialised vision system from scratch—a process that typically requires months of development—Orakl adapted the pre-trained model to their specific organoid imagery.
Meanwhile, Mendel AI tackled another critical healthcare challenge: the inefficient matching of patients to clinical trials, a process that traditionally takes hundreds of days and causes approximately 80% of trials to miss enrollment targets. Mendel's flagship product, Hypercube, addresses this by combining Meta's open-source Llama large language models with a specialised clinical hypergraph to create a secure, searchable knowledge base.
Mendel's implementation journey began with fine-tuning Llama 2 to create a natural language interface that could translate user queries into their symbolic query language. The company then expanded its AI capabilities by continuously pre-training both 8B and 70B parameter versions of Llama 3, creating a healthcare-specific foundation model.
A key advantage for both companies was the open-source nature of Meta's models, which eliminated the need for massive upfront investment in AI development while allowing them to maintain control of sensitive healthcare data. Unlike cloud-based proprietary AI systems that might require sending data to third parties, implementations using Llama and DINOv2 allow healthcare companies to keep their data on their own infrastructure, addressing strict regulatory requirements.
Both implementations have delivered transformative results: Orakl Oncology developed their platform in months rather than years, while Mendel AI reduced the patient matching process from hundreds of days to just one day. These accelerations dramatically reduce costs for pharmaceutical companies and research institutions while potentially bringing life-saving treatments to market faster.
Looking ahead, both companies plan to enhance their platforms further—Orakl by extending their quantitative analysis capabilities to more complex organoid models, and Mendel by implementing the multimodal capabilities of Llama 3.2 to process not just text but also images and other clinical data formats.
These AI implementations have transformed critical healthcare processes across the drug development value chain. Orakl's 26.8% accuracy improvement translates directly to more reliable predictions of treatment efficacy, potentially saving pharmaceutical partners millions in avoided failed trials. Mendel's reduced matching time from months to one day allows trials to begin treatment phases sooner, potentially reducing time-to-market for new therapies. Both companies have established significant competitive advantages through superior performance compared to traditional approaches. Their ability to operate on clients' own infrastructure addresses key regulatory and data privacy concerns in the highly regulated healthcare industry.