SP 2

Mining spatial and functional immune cell patterns to develop and clinically validate novel prognostic tissue markers

The goal of this consortium is to investigate and exploit the full prognostic potential of inflammation reactions related to tumor cells in hereditary breast cancer as well as for renal allografts after transplantation and make it available for translational research. In subproject SP2 modules for robust image and data analysis are developed so that the spatial context of immune-cell distributions are characterized. These modules are based on work performed within the first funding period. Analysis results are stored in a unified way in databases to make them available for further use of project partners. Based on these data and by developing and applying methods of data mining and systems immunology diagnostic algorithms with high prognostic or predictive power are created.

Figure 1: Definiens VeriTrova™ software features an automatically aligned display of multiple sections of a tissue block, which enables the analysis of cell co-occurrence patterns. The image shows a CD3/CD20 dual-stained section in the main window used to investigate distributions of T and B cells. In the smaller windows corresponding regions of other stains are displayed. Using this platform also regions of interest can be annotated. These are either used as training material for machine learning procedures or quantitatively enriched to form a data base for data mining approaches. In the example here glomerula structures are marked red. The status of these structures is relevant for the cleaning capability – and thus overall function – of the kidney  

Using an iterative process and in close collaboration with project partners (especially with SP4) we will discover and validate diagnostically relevant “Phenes”. “Phenes” are mathematically descriptions of spatial and functional entities derived from immune-cell distributions in tissue. Due to the possibly large number of Phene candidates but a restricted availability of patient data, new and robust data mining algorithms have to be developed. Those ensure the successful translation of our findings to prospective studies.

Figure 2: Workflow for heat map generation. Results of detailed automated image analyses (here: identification of cells) are stored in Definiens Result Container files (DRCs).These are based on the HDF5 standard and constitute a data repository to store image analysis results which are also accessible to project partners for further processing.


Keywords: Image, analysis, Tissue Diagnostics, Image Mining