Foundation models represent a new generation of AI, equipped with a vast knowledge base that enables them to tackle a wide range of complex problems. With the Foundation Model Initiative (HFMI), Helmholtz is already supporting four pilot projects on the topics of weather & climate, CO2 cycles, radiology, and photovoltaic materials. In addition, the community is investing in the necessary computing infrastructure and has established a Synergy Unit, which conducts research on overarching topics. The initiative's three-year projects aim to develop fully functional models.
VirtualCell: A Foundation Model for Cell and Tissue State Across Health and Disease
VirtualCell aims to realize the long-standing vision to create a digital twin-like replica of a cell, enabling insights into the complex cellular processes that occur in health and disease. By leveraging recent advances in high-throughput genomic sequencing, imaging, and generative AI, the project will develop a multimodal foundation model trained on extensive single-cell multi-omics and spatial data. This model will provide a comprehensive representation of cellular states and interactions, which will be fine-tuned on disease samples from biomedical partners for various clinical applications.
“The implications for medical research are significant,” says Prof. Fabian Theis, Director at Helmholtz Munich’s Computational Health Center and leader of the VirtualCell team. “Most disease happens on the cellular level. Through this advanced cellular modeling, VirtualCell is poised to drive breakthroughs in disease pathology, drug design, and patient stratification, ultimately driving advancements in biotechnology applications and our understanding of complex diseases.”
VirtualCell is a joint initiative between Helmholtz Munich, the Max Delbrück Center, Forschungszentrum Jülich, and Nvidia.
PROFOUND: Revolutionizing Protein Design
Proteins are the building blocks of life, but their dynamic nature – constantly changing shape to perform biological tasks – poses a challenge for current AI models like AlphaFold, which predict only static structures. PROFOUND aims to overcome this limitation by integrating large-scale molecular dynamics data to predict the dynamic behaviors of proteins. This approach could revolutionize the design of proteins that not only perform specific functions but also adapt and evolve over time.
“By modeling proteins in motion, PROFOUND opens the door to designing dynamic enzymes and programmable molecular machines,” explains Prof. Stefan Bauer, Senior PI at Helmholtz AI, who leads the AI part of the PROFOUND project. The project is in close collaboration with domain experts including Dr. Christoph Gruber and Dr. Florian Giesert who highlight its potential impact: “This could lead to breakthroughs in smart therapeutics, sustainable materials, and next-generation biotechnologies.”
PROFOUND is a joint project involving Forschungszentrum Jülich (lead), CASUS, Helmholtz-Zentrum Berlin and Helmholtz Munich.
Pushing the Frontiers of Medical Research with Foundation Models
With VirtualCell and PROFOUND, Helmholtz Munich and its Computational Health Center is leading the charge in applying foundation models to medical research, a crucial step toward unlocking the complexities of biological systems. These models, capable of integrating vast amounts of diverse data, offer unprecedented opportunities to enhance our understanding of disease mechanisms and cellular behavior. The Helmholtz Association’s support of these projects underscores the growing importance of foundation models in advancing precision medicine and biotechnology. By pushing the boundaries of what is possible in protein design and single-cell biology, Helmholtz Munich is paving the way for more personalized and effective treatments, driving progress in both healthcare and biomedical research.
