AI foundation models: millions in funding for three further pilot projects
The so-called Foundation Models are a new generation of AI models that have a broad knowledge base and are therefore able to solve a range of complex problems. They are significantly more powerful and flexible than conventional AI models and thus hold enormous potential for modern, data-driven science. They can become powerful tools that answer a variety of research questions.
Helmholtz is a pioneer in this field and is already funding four pilot projects in the areas of weather and climate, CO2 cycles, radiology and photovoltaic materials with the Foundation Model Initiative (HFMI). In addition, the association has set up a Synergy Unit to research overarching topics and invest in the necessary computing infrastructure. The aim of the three-year Initiative is to develop fully functional models. Now, a panel of experts has selected three further projects for funding.
AqQua
Life in the water plays a crucial role in the Earth's climate. Plankton in particular binds large amounts of carbon from the atmosphere. Climate change is altering plankton ecosystems and affecting carbon export and food resources in the ocean. Despite its enormous importance, we know surprisingly little about the abundance of most plankton species. There are also major uncertainties in estimates of global marine carbon export. Every day, researchers around the globe take millions of plankton images using a variety of instruments. Distributed pelagic imaging enables comprehensive monitoring of aquatic life down to the bottom of the deep sea. AqQua will combine billions of such images to develop the first fundamental pelagic imaging model that can be used for global monitoring of plankton biodiversity, ecosystem health and carbon flux. In this way, we will support decision-making in times of global change, particularly with regard to new technologies for removing carbon dioxide from the atmosphere.
Participating Centers: Forschungszentrum Jülich, GEOMAR, HEREON, Max Delbrück Center
PROFOUND
Proteins are the fundamental molecules of life. They are building materials and chemical tools that control the processes inside living things. It is the way they interact and function that makes all living systems so complex and fascinating. The ability to design proteins has enormous potential, for basic biology many applied areas and medicine. This year's Nobel Prize in Chemistry was awarded to researchers who have pioneered this field. Today, it is possible to predict the three-dimensional structure of the protein that encodes a gene sequence using AI. PROFOUND will revolutionize protein design by overcoming a major hurdle faced by current AI models such as AlphaFold: they are limited to static protein structures. In reality, proteins are like nanomachines that constantly change shape to perform their biological work. PROFOUND aims to capture these motions. PROFOUND will leverage large-scale molecular dynamics data to create an AI model that can predict these dynamic behaviors. This approach will allow us to design proteins that not only perform specific tasks but also adapt over time. We can envision breakthroughs such as dynamic enzymes and programmable molecular machines that could lead to innovations in smart therapeutics, sustainable materials, and next-generation biotechnologies.
Participating Centers: Forschungszentrum Jülich, Helmholtz Munich, Helmholtz-Zentrum Berlin, Helmholtz-Zentrum Dresden-Rossendorf
VirtualCell
It is a long-standing vision to create a digital twin like replica of a cell. This would enable not only insight into the plethora of ongoing complex cellular processes but also effect prediction under perturbations such as disease or drugs. Building upon recent advances in high-throughput genome sequencing and imaging combined with generative AI, the VirtualCell project aims to tackle this ambitious challenge. The project will develop a multimodal baseline model trained on extensive single-cell multi-omics and spatial data to provide a comprehensive representation of cellular states and interactions. The model will be applied to a range of novel clinical tasks by adapting it to disease samples from biomedical partners. By advancing cellular modeling, VirtualCell is poised to achieve breakthroughs in disease pathology, drug development, and patient stratification, and to significantly improve biomedical research and healthcare outcomes.
Participating Centers: Forschungszentrum Jülich, Helmholtz Munich, Max Delbrück Center
The Helmholtz Association is ideally placed to develop such future-oriented applications: a wealth of data, powerful supercomputers on which the models can be trained, and in-depth expertise in the field of artificial intelligence.
