Each project will receive up to €150,000 from the Helmholtz Initiative and Networking Fund. The research teams will collaborate closely with Helmholtz AI and Helmholtz Imaging to establish best practices in metadata management, reproducible workflows, and open-access datasets. Starting in January 2026, the projects will take part in workshops to share knowledge and foster the emerging AI benchmarking community. Together, these projects demonstrate Helmholtz Munich’s contribution to shaping the next generation of trustworthy AI across scientific domains.
The three selected projects in overview:
SCHEMA Develops AI-Ready Dataset to Predict Cancer Metastasis
SCHEMA brings together Helmholtz Munich and partner institutions to create the largest single-cell, spatial dataset for predicting metastasis in lung, colon and breast cancer. By combining public data with newly generated tumor profiles, the project provides a benchmark for AI scientists to develop models that forecast which tumors are likely to spread. The dataset will be openly accessible, harmonized for reproducibility, and paired with community workshops to foster collaboration across AI, biology, and clinical research. SCHEMA aims to accelerate discovery of predictive biomarkers, guide personalized treatment, and ultimately support new strategies for tackling cancer metastasis.
Project leads: Dr. Malte Lücken and Prof. Markus E. Diefenbacher
Learn more: SCHEMA – profiling Spatial Cancer HEterogeneity across modalities to benchmark Metastasis risk prediction – helmholtz-imaging.de
TIMELY Sets Benchmark for Multimodal Biological Time-Series Data
TIMELY unites Helmholtz Munich and partners to create the first comprehensive benchmark for biological time-series data across neuroscience, cell biology, behavior, and ecology. By integrating diverse datasets (from cortical activity and brain organoid recordings to live-cell imaging and mouse behavior) TIMELY provides standardized, high-quality resources for developing AI and statistical models that capture complex biological dynamics. The open-access benchmark, paired with reference code and community workshops, promotes reproducibility, cross-disciplinary collaboration, and robust evaluation of models. TIMELY aims to accelerate discoveries in biomedicine and neuroscience and provide a foundation for AI-driven insights into dynamic biological systems.
Project lead: Dr. Steffen Schneider
Learn more: TIMELY: Time-series Integration across Modalities for Evaluation of Latent DYnamics – helmholtz-imaging.de
UQOB Develops Multi-Rater Benchmark for Reliable Object Detection in Organoid Imaging
UQOB develops the first benchmark dataset for object detection and uncertainty quantification in microscopy images of alveolar and bronchiolar organoids from mouse and human lungs, complemented by colon organoid data. The dataset will include over 800 high-resolution images and 120,000 annotated organoids, each image will be labeled by multiple experts to capture biological diversity and annotation variability. By combining deep phenotype annotation with standardized protocols and open-source baselines, UQOB enables the optimization of organoid detection and the development of more trustworthy AI models. The project will culminate in a public Kaggle challenge, fostering innovation in uncertainty-aware AI for biomedical imaging.
Project Lead: Dr. Marie Piraud
Learn more: UQOB – Uncertainty Quantification in Object-detection Benchmark – helmholtz-imaging.de
