We are a research group at the Heidelberg University and Heidelberg University Hospital studying how the spatial organization of tissue shapes cell state, disease progression and response to therapy. We develop interpretable AI/ML and optimization methods that turn imaging, single-cell and spatial omics data into mechanistic hypotheses and clinical prediction.

OWe work at the intersection of AI and biomedicine to address a fundamental challenge in biology and medicine: understanding the principles governing tissue organization and plasticity. We develop interpretable, scalable machine learning and optimization approaches that turn the complexity of single-cell and spatial omics data into mechanistic insight and clinical prediction.

• Cell state depends on tissue context. The molecular state of a cell is predictable in part from its surroundings, and the contribution of context can be separated from the contribution of the cell itself. We build multiview and graph-based models that keep this decomposition explicit rather than absorbing it into a single latent space, across transcript, protein and imaging modalities.

• Contextual programs are local and persistent. Patterns of intra- and intercellular relationship recur at consistent spatial scales across samples and conditions, which makes them learnable from sample-level labels without cell-level annotation. We develop compressed representations that preserve this locality, so that what the model used remains recoverable.

• Tissue organization changes in structured ways.** Spatial programs are reconfigured across disease stages and under therapeutic pressure, and those trajectories are constrained rather than arbitrary. We formulate this as an alignment and transport problem over tissue geometry, to reconstruct how organization evolves and to generate hypotheses about adaptation, immune evasion and resistance.

• Tissue geometry carries clinical signal beyond molecular composition. Spatial arrangement predicts progression and therapeutic response where composition alone does not. We build explainable models that expose which spatial patterns drive a given prediction and distil them into compact, testable descriptions of tissue for patient stratification.

We value collaborations with clinical, experimental biology groups and groups working on the development of novel methods for the acquisition of single-cell spatial omics data. We welcome synergistic collaborations with computational groups towards the construction of more robust theoretical and computational frameworks for the analysis of all aspects of biomedical data and beyond.