French artificial intelligence startup Bioptomus, pioneers of the biology research model family H-Optomus, today announced the availability of the company’s next-gen clinical biology AI model, H-Optomus-1, and the world’s first universal multimodal AI foundation model for research, M-Optimus.

An iteration on the company’s first model, H-Optimus-0, released in late 2024, H-Optimus-1 is currently the leading foundation model for histology, the microscopic study of biological tissues to understand their structure. Principally, histology reveals the microscopic anatomy of how cells organize into tissues and organs.

The company said H-Optimus-1 represents a breakthrough in tissue analysis to assist with the identification of disease and other pathologies. It tops the comprehensive multi-task, multi-organ benchmarks, including PathBench from the Hong Kong University of Science and Technology and the HEST benchmark from Harvard.

The launch of M-Optimus heralds the unveiling of a unifying foundation model that uses the increasing body of multimodal biomedical data to enable biological discovery and development.

It’s capable of modeling robust representations of cells, tissues and patients across diseases and populations, to provide a universal framework for predicting clinical outcomes through “digital twins.” These are essentially virtual biological avatars that can be used to forecast disease progression and individual treatment response.

By providing an artificial simulation basis for tissues and disease progression, it can guide therapeutic decisions and scale computer-driven clinical trials.

“With M-Optimus, we have successfully assembled the first critical components of our journey to crack the code of biology by combining multiple modalities at scale.,” said Jean-Philippe Vert, co-founder and chief executive of Bioptimus.

Bioptimus described the new model as a universal model trained on the world’s largest proprietary datasets. The training data incorporates curated information from millions of patients, more than 50 organ types and hundreds of medical centers — this broad reach provides the model profound biological diversity and depth.

M-Optimus helps solve pressing issues in clinical science regarding the prediction of gene expression, treatment responses and outcomes using direct multimodal and histological data. The company said it opens opportunities to accelerate drug design through predictive modeling and simulation of patient response.

The ability to generate digital twins of tissues, patients and populations allows the model to assist in large-scale clinical trials, drastically reducing the time and cost for testing novel treatments.

The medical research industry has long faced a gap between molecular-level biology and clinical decisions, since transitioning from the lab to the real world can be a fraught process. With realistic modeling, lab work and clinical trials can be tailored to core opportunities and cover edge cases without overlap or “shooting in the dark.”

Researchers can also train their own versions by fine-tuning with proprietary data, preserving private corporate information while retaining the model’s universal understanding capabilities.

Alongside the launch of M-Optimus, the company announced that H-Optimus-1 will be available with broad access on Amazon Web Services Inc.’s SageMaker, the cloud company’s fully managed environment for building, training and deploying AI models.

Image: SiliconANGLE/OpenAI ChatGPT