Foundation Models
At Helmholtz Munich, we develop and use foundation models – advanced artificial intelligence (AI) systems – to improve health research and medical care. These models analyze vast amounts of data to detect patterns, helping experts identify diseases earlier, personalize treatments, and enhance healthcare. By combining genetic, environmental, and medical information, foundation models provide deeper insights into how diseases develop and how to treat them more effectively. Our goal is to harness these technologies to advance prevention, treatment, and overall well-being for everyone.
At Helmholtz Munich, we develop and use foundation models – advanced artificial intelligence (AI) systems – to improve health research and medical care. These models analyze vast amounts of data to detect patterns, helping experts identify diseases earlier, personalize treatments, and enhance healthcare. By combining genetic, environmental, and medical information, foundation models have the potential to provide deeper insights into how diseases develop and how to treat them more effectively. Our goal is to harness these technologies to advance prevention, treatment, and overall well-being for everyone.
What Are Foundation Models?
Foundation models are advanced AI systems trained on massive amounts of data to perform a wide range of tasks. Unlike traditional AI models designed for specific purposes, foundation models learn general patterns from diverse data sources and can be adapted for various applications, such as language processing, image recognition, and scientific research.
In health and science, foundation models help analyze complex biological, medical, and environmental data. They can identify disease patterns, personalize treatments, and accelerate discoveries in areas like drug development and genetics. Their ability to process vast information makes them powerful tools for advancing research and improving decision-making across many fields.
Foundation models are unlocking the potential of complex biological data, paving the way for smarter, more personalized treatments.
Hot Topics
Imagine having a map of every cell type in the human body - what it looks like, what it does, and how it changes in disease. That’s the goal of a "cell atlas." Researchers at Helmholtz Munich are at the forefront of this effort. They are pioneering the use of Foundation Models to analyze millions of single-cell datasets. These AI models can spot patterns, predict how cells behave, and even uncover hidden relationships between genes.
Why it matters? This could help researchers understand diseases earlier, develop new treatments, and even personalize medicine based on a person’s unique cellular fingerprint.
Read the story about Fabian Theis' research: A Journey into the Secrets of Human Cells
Foundation Models are powerful AI systems trained on vast amounts of data from across the internet. But with great power comes great responsibility. How can we make sure these models are fair, respect privacy, and avoid bias? Our researchers are working on ways to understand how these models make decisions - and how to make those decisions more transparent, explainable, and ethical.
Why it matters? As AI systems are used in healthcare, education, and public policy, ethical use becomes a top priority. We need to ensure AI is built and used in a way that benefits everyone.
Read the interview with Julia Schnabel: FUTURE-AI: Making AI in Healthcare Trustworthy!
Read the news about FUTURE-AI: International Experts Establish FUTURE-AI Guidelines for Trustworthy Healthcare AI
Read the interview with Marcel Binz: AI in Science: Ethical and Practical Challenges
Read the Feature Publication: Addressing Bias in Machine Learning for Equitable Healthcare
Helmholtz Munich researchers are exploring how Foundation Models can help us better understand human learning and decision-making. By combining behavioral data, language, and visual information, their research aims to model how people reason, learn from feedback, and make choices in complex environments.
Why it matters? These insights can support the development of AI systems that interact more naturally with humans - whether in education, mental health, or personalized decision support tools.
Read the interview with Eric Schulz: "Teaching AI Psychological Skills for Better Diagnosis and Therapies"
Read the interview with Zeynep Akata: “Extending the Limits of Our Curiosity With the Help of Technological Tools.”
At Helmholtz Munich, researchers are developing Foundation Models that can analyze digital tissue images with remarkable accuracy. Trained on thousands of histopathology slides, these AI tools help pathologists detect cancer, inflammation, and other changes more quickly and consistently.
Why it matters? Rather than replacing doctors, this technology supports them - speeding up diagnoses, reducing errors, and ensuring expert-level analysis is available even in under-resourced settings.
Read the story about Carsten Marr's research: AI Transforms Blood Disease Diagnosis
Read the Featured Publication about Tingying Peng's research: New AI Model Enhances Speed and Accuracy in Medical Diagnoses
Read the interview with Valentin Koch & Sophia Wagner: DinoBloom: A Foundation Model for Generalizable Cell Embeddings in Hematology
At Helmholtz Munich, researchers are using Foundation Models to speed up drug discovery. By analyzing chemical structures and biological data, these AI systems predict which molecules could become effective medicines.
Why it matters? AI helps scientists identify promising drug candidates faster, reducing the time and cost of developing life-saving treatments.
Read the news about Ewa Szczurek's research: A New Generation of Antimicrobial Peptides
Read the interview with Roberto Olayo Alarcon: MolE: Pre-trained molecular representations enable antimicrobial discovery
Researchers at Helmholtz Munich are leveraging Foundation Models to analyze complex genetic data. By decoding patterns in DNA, these AI models help identify genetic mutations, understand disease mechanisms, and predict individual health risks.
Why it matters? This approach in genetic diagnostics can lead to earlier, more accurate diagnoses and pave the way for precision medicine.
Read the news about Julien Gagneur's research: Better Search for the Cause of Hereditary Diseases
Read the news about Annalisa Marsico's research: More than the sum of mutations – 165 new cancer genes identified with the help of machine learning
Meet the Researchers
Selected Foundation Model Publications
Selected Foundation Model Publications
Luca M. Schulze Buschoff, Elif Akata, Matthias Bethge & Eric Schulz. 2025
Visual cognition in multimodal large language models. Nature Machine Intelligence
Mohammad Lotfollahi, Yuhan Hao, Fabian J. Theis, Rahul Satija. 2024
The future of rapid and automated single-cell data analysis using reference mapping. Cell
Anna C. Schaar, Alejandro Tejada-Lapuerta, Giovanni Palla, Robert Gutgesell, Lennard Halle, Mariia Minaeva, Larsen Vornholz, Leander Dony, Francesca Drummer, Mojtaba Bahrami, Fabian J. Theis. 2024.
Nicheformer: a foundation model for single-cell and spatial omics. bioRxiv
Alexander Karollus, Johannes Hingerl, Dennis Gankin, Martin Grosshauser, Kristian Klemon & Julien Gagneur. 2024
Species-aware DNA language models capture regulatory elements and their evolution. Genome Biology
Manuel Tran, Paul Schmidle, Sophia J. Wagner, Valentin Koch, Valerio Lupperger, Annette Feuchtinger, Alexander Böhner, Robert Kaczmarczyk, Tilo Biedermann, Kilian Eyerich, Stephan A. Braun, Tingying Peng, Carsten Marr. 2024.
Generating highly accurate pathology reports from gigapixel whole slide images with HistoGPT. medRxiv
Roberto Olayo-Alarcon, Martin K. Amstalden, Annamaria Zannoni, Medina Bajramovic, Cynthia M. Sharma, Ana Rita Brochado, Mina Rezaei, Christian L. Müller. 2024.
Pre-trained molecular representations enable antimicrobial discovery. bioRxiv
Sergey Vilov, Matthias Heinig. 2024.
Investigating the performance of foundation models on human 3’UTR sequences. bioRxiv
Marcel Binz, Stephan Alaniz, Adina Roskies, Balazs Aczel, Carl T. Bergstrom, Colin Allen, Daniel Schad, Dirk Wulff, Jevin D. West, Qiong Zhang, Richard M. Shiffrin, Samuel J. Gershman, Ven Popov, Emily M. Bender, Marco Marelli, Matthew M. Botvinick, Zeynep Akata, Eric Schulz. 2023.
How should the advent of large language models affect the practice of science?. arXiv
Stéphane d'Ascoli, Sören Becker, Alexander Mathis, Philippe Schwaller, Niki Kilbertus. 2023.
ODEFormer: Symbolic Regression of Dynamical Systems with Transformers. arXiv
Media Contact
Talk to our press office
